CN102661759B - Method for identifying and predicting nonlinear multivariable key parameters of circulating fluidized bed boiler - Google Patents

Abstract

The invention discloses a method for identifying and predicting nonlinear multivariable key parameters of a circulating fluidized bed boiler, which comprises the following steps of: 1, real-time detection and synchronous transmission of factor signals affecting combustion efficiency of circulating fluidized bed boiler; 2, signal sampling and pretreatment; 3, identification of the nonlinear multivariable key parameters of the circulating fluidized bed boiler; 4 real-time prediction of the nonlinear multivariable key parameters of the circulating fluidized bed boiler; 5, repetition of the step 3 and step 4 to identify all parameters of tested key variables and predict the real-time predicted values of all key parameters of the circulating fluidized bed boiler; and 6, output of results. The method is reasonably designed and convenient to implement; the method can predict the real-time key parameters affecting the stable operation of the circulating fluidized bed boiler with high accuracy, high efficiency, less time and less labor, while making real-time fault diagnosis; therefore, the method is worthy to be used widely.

Description

Identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter

Technical field

The present invention relates to gangue boiler of power plant Parameter identification technical field, especially relate to a kind of identification and Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter.

Background technology

China is coal big country, and 2011 produce approximately 35.7 hundred million tons, raw coal per year, wherein 15%～20% colm such as gangue, coal slime for discharge, and gangue combustion power generation is the form of a kind of energy-saving and emission-reduction of thermal power generation, there is approximately nearly 200 of gangue power plant in coal enterprise at present.Circulating Fluidized Bed Boiler is the key equipment of coal gangue power generation factory safe power generation, the burning efficiency of Circulating Fluidized Bed Boiler will be subject to the impact of the factors such as primary air fan, overfire air fan, returning charge blower fan, induced draft fan, gangue flow, and the variation of parameters all has nonlinear feature.International and domestic expert has carried out a large amount of research for the matching problem of fault detect, power plant's induced draft fan and the pressure fan of waste material generating, Fan for Boiler in Power Plant, but does not see the research of the Nonlinear Multivariable about coal gangue power generation factory Circulating Fluidized Bed Boiler is related to.Furnace pressure to boiler, main steam pressure are carried out and are controlled, and can guarantee the stable operation of power plant boiler burning.Production technology according to gangue storehouse to boiler, the fuel of gangue power plant is through the processing of disintegrating machine and vibratory screening apparatus, coal particle size is less than 13mm, in conjunction with the actual conditions of gangue power plant, returning charge blower fan is that the larger particles fuel that fully burning separates through separation vessel is sent into burner hearth is again combustion-supporting, iterative cycles burning, improves burning efficiency, the oxygen level O2 of the flue gas of gangue flow Im and smoke evacuation is the principal element that affects boiler combustion efficiency, therefore with gangue flow Im, the crucial controlled variable that the flue gas oxygen content O2 of the pressure P of returning charge blower fan and smoke evacuation is Circulating Fluidized Bed Boiler, these crucial controlled variables are subject to again the impact of other variable, for example crucial measured variable gangue flow Im will be subject to the pressure of returning charge blower fan, the electric current of returning charge blower fan, flue gas oxygen content O2, the electric current of induced draft fan, primary air fan electric current, the impact of overfire air fan electric current, be complicated nonlinear relation, if can be to these crucial controlled variable identifications, and carry out real-time estimate, can be raising boiler combustion efficiency, turbine generator stable operation is laid the foundation.

Conventional parameter identification method has the discrimination method of classic identification method, least square method, maximum-likelihood method, gradient calibration identification and model reference adaptive based on related function, but these discrimination methods are all for Linear Time-Invariant System, the identification of nonlinear system generally adopts neural network, the identification of Volterra progression also can be for nonlinear identification, but to first find the corresponding relation of Volterra sum of series wide area frequency response (GFRF), then ask the frequency response of the kernel of Volterra with the GFRF recursive algorithm of nonlinear system, calculation of complex.Have not yet to see N the parameter identification method about the crucial measured variable of Circulating Fluidized Bed Boiler Nonlinear Multivariable.

Summary of the invention

Technical matters to be solved by this invention is for above-mentioned deficiency of the prior art, a kind of identification and Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter are provided, it is reasonable in design, it is convenient to realize, identification precision is high and save time, can carry out real-time estimate and fault diagnosis, application value is high.

For solving the problems of the technologies described above, the technical solution used in the present invention is: a kind of identification and Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, is characterized in that the method comprises the following steps:

Step 1, on affect burning in circulating fluid bed boiler efficiency multiple factor signals real-time detection and synchronously upload: by multiple sensors, the multiple factor signals that affect burning in circulating fluid bed boiler efficiency are detected in real time, and real-time detected signal are synchronously uploaded to data collecting card;

Step 2, signal sampling and pre-service: data collecting card carries out L sampling to the detected signal of multiple sensors and correspondingly amplifies, after filtering and A/D conversion process, obtains L × N dimension matrix samples signal x _kiand be synchronously uploaded to processor and carry out record; Wherein, k=1,2 ..., L, i=1,2 ..., N, the value of N equates with the quantity of sensor and N is greater than 2 natural number, L is greater than 20 natural number;

The identification of step 3, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the L obtaining in step 2 × N is tieed up to matrix samples signal x by processor _kicarry out analyzing and processing and identification, draw N parameter of any one tested key variables of Circulating Fluidized Bed Boiler, its analyzing and processing and identification process are as follows:

301, processor is to L × N dimension matrix samples signal x _kicarry out analyzing and processing, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz, its analyzing and processing process is as follows:

3011, select the RBF kernel function of SVM: for L sample signal of any one crucial measured variable of Circulating Fluidized Bed Boiler, i.e. L × N ties up matrix samples signal x _kiany one row column vector x _i=[x _1ix _2ix _li] ^t, wherein, i represents columns and i=1～N, the RBF kernel function of its SVM is:

k(x _t,x _j)=exp(-||x _t-x _j|| ²/2s ²) （1-1）

Wherein, x _trepresent that any one crucial measured variable of Circulating Fluidized Bed Boiler is at the measured value in t moment, x _tvalue be column vector x _iin any one value; x _jrepresent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler, x _jvalue be column vector x _ithe average of middle all values; S represents the number of getting SVM;

3012, processor is according to the RBF kernel function of selecting in 3011, by L × N dimension matrix samples signal x _kifrom N dimension real domain space R ⁿbe mapped to higher-dimension Hilbert space, in conjunction with batch processing least square method, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz:

$H_{\mathrm{Lz}}={\left[\begin{array}{ccccc}{k}_z(x_{1z},x_{\mathrm{zj}})& k_2(x_{12},x_{2j})& k_3(x_{13},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_z(x_{1N},x_{\mathrm{Nj}})\\ k_z(x_{2z},x_{\mathrm{zj}})& k_2(x_{22},x_{2j})& k_3(x_{23},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{2N},x_{\mathrm{Nj}})\\ k_z(x_{3z},x_{\mathrm{zj}})& k_2(x_{32},x_{2j})& k_3(x_{33},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{3N},x_{\mathrm{Nj}})\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_z(x_{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">L</figure-callout>}2},x_{\mathrm{zj}})& k_2(x_{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">L</figure-callout>}2},x_{2j})& k_3(x_{\mathrm{<figure-callout\; id="3"\; label="L"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">L</figure-callout>}3},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{\mathrm{LN}},x_{\mathrm{Nj}})\end{array}\right]}^T---(1-2)$

Wherein, multidimensional deviation sample kernel function collection H _lzthe first row row vector k _z(x _1z, x _zj), k _z(x _2z, x _zj), k _z(x _3z, x _zj) ..., k _z(x _lz, x _zj) be L sample kernel function of crucial measured variable, multidimensional deviation sample kernel function collection H _lzi every trade vector k _i(x _1i, x _ij), k _i(x _2i, x _ij), k _i(x _ki, x _ij) ..., k _i(x _li, x _ij) be L sample kernel function of the relevant measured variable of i the crucial measured variable of impact, i=2～N;

302, processor is according to formula

$H_{\mathrm{hz}}={\left(H_{\mathrm{Lz}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H}_{\mathrm{Lz}}^T=\left[\begin{array}{ccccc}{k}_{11}& k_{12}& k_{13}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{1L}\\ k_{21}& k_{22}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{2L}\\ k_{31}& k_{32}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{3L}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_{N1}& k_{N2}& k_{\mathrm{<figure-callout\; id="3"\; label="N"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">N</figure-callout>}3}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{\mathrm{NL}}\end{array}\right]---(1-3)$

By multidimensional deviation sample kernel function collection H _lzbe transformed to synthetic nucleus function sample matrix H _hz;

303, processor builds the random deviation dynamic model of any one crucial measured variable of Circulating Fluidized Bed Boiler in higher-dimension Hilbert space:

$z_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})---(1-4)$

And draw random deviation dynamic model parameter set according to batch processing least square method and the identification of least disadvantage function principle $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T,$ Wherein, a representative is by the parameter of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, b _irepresent that i impact is by the parameter of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification,

represent the identifier of a, represent b _iidentifier, k _i(x _ti, x _ij) represent that i impact is by the kernel function of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, k _i(x _ti, x _ij) value be multidimensional deviation sample kernel function collection H _lzi every trade vector in any one value; Its identification draws random deviation dynamic model parameter set

process as follows:

3031, processor builds the identifier prediction type of any one crucial measured variable of Circulating Fluidized Bed Boiler according to batch processing least square method:

x _kz(t+1)=h ^Tθ=ax _kz(t)+b ₂x _k2(t)+b ₃x _k3(t)+…+b _Nx _kN(t) （1-5）

Wherein, x _kz(t) represent that any one crucial measured variable of Circulating Fluidized Bed Boiler is in the deviate in t moment, any one crucial measured variable of Circulating Fluidized Bed Boiler is in the difference of the measured value in t moment and all measured value averages of any one crucial measured variable of Circulating Fluidized Bed Boiler; h ^t=[x _kz(t) x _k2(t) x _k3(t) ... x _kN(t)] be L × N dimension matrix samples signal x _kii every trade vector, θ represents the random deviation dynamic model parameter set that identification draws

true value, and θ=[a b ₂b ₃b _n] ^t;

The measured value z (k) and its identifier x of any one crucial measured variable that 3032, the loss function of processor definition identification is Circulating Fluidized Bed Boiler _kz(t+1) Quadratic Function Optimization of the error between, is expressed as:

$J\left(\mathrm{\&theta;}\right)=\frac{1}{2}\underset{k=1}{\overset{L}{\mathrm{\&Sigma;}}}{[z\left(k\right)-h^T\mathrm{\&theta;}]}^2---(1-6)$

Make loss function minimum, meet:

$\left\{\begin{array}{c}\frac{\&PartialD;J\left(\mathrm{\&theta;}\right)}{\&PartialD;\mathrm{\&theta;}}=0\\ \frac{{\&PartialD;}^{2}J\left(\mathrm{\&theta;}\right)}{{\&PartialD;\mathrm{\&theta;}}^2}>0\end{array}\right.---(1-7)$

Obtain the identifier of θ:

$\widehat{\mathrm{\&theta;}}={\left({H_{\mathrm{Lz}}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H_{\mathrm{Lz}}}^T{z}_L=H_{\mathrm{hz}}{z}_L---(1-8)$

Wherein, z _lrepresent in multiple described sensors any one L sampling output valve, the i.e. L of any one crucial measured variable of a Circulating Fluidized Bed Boiler measured value;

3033, processor is according to formula

$\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T=H_{\mathrm{hz}}{z}_L---(1-9)$

Identification draws the random deviation dynamic model parameter set of any one crucial measured variable of Circulating Fluidized Bed Boiler

be N parameter of identification any one key variables of having drawn Circulating Fluidized Bed Boiler;

The real-time estimate of step 4, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the random deviation dynamic model of any one crucial measured variable of the Circulating Fluidized Bed Boiler of processor based on building in 303, builds the hierarchical prediction model of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter:

$\left\{\begin{array}{c}{z}_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})\\ z(k+1)=z_0\left(k\right)+z_e\left(k\right)\end{array}\right.---(1-10)$

And according to the random deviation dynamic model parameter set of any one crucial measured variable of the Circulating Fluidized Bed Boiler that in hierarchical prediction model and 303, identification draws

dope the real-time estimate value z (k+1) of any one crucial measured variable of Circulating Fluidized Bed Boiler; Wherein, z ₀(k) represent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler;

Step 5, repeat step 3 and step 4, until identification draws all parameters of all crucial measured variables of Circulating Fluidized Bed Boiler, and dope the real-time estimate value of all crucial measured variables of Circulating Fluidized Bed Boiler;

Step 6, result are synchronously exported: in step 3, carry out in the identification process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and in step 4, carry out in the real-time estimate process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, processor by the display that joins with it in the identification result to the signal processing in step 3 and Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and step 4 predicting the outcome of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter carry out simultaneous display.

Identification and the Forecasting Methodology of above-mentioned Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, it is characterized in that: the quantity of sensor described in step 1 and step 2 is 7,7 described sensors are respectively gangue flow sensor, induced draft fan current transformer, returning charge blower fan current transformer, returning charge blower pressure sensor, primary air fan current transformer, overfire air fan current transformer (6) and flue gas oxygen content sensor.

Identification and the Forecasting Methodology of above-mentioned Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, is characterized in that: all crucial measured variable of Circulating Fluidized Bed Boiler described in step 5 is gangue flow, returning charge blower pressure and flue gas oxygen content.

Identification and the Forecasting Methodology of above-mentioned Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, is characterized in that: described processor is industrial control computer.

Identification and the Forecasting Methodology of above-mentioned Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, it is characterized in that: while carrying out the identification of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 3, and while carrying out the real-time estimate of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 4, described industrial control computer is realized by MATLAB, Visual C and configuration software.

Identification and the Forecasting Methodology of above-mentioned Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, is characterized in that: in step 2 and step 3, the value of L is 20～50; In 3011, the value of s is 10～50.

The present invention compared with prior art has the following advantages:

1, the present invention is reasonable in design, and it is convenient to realize.

2, the present invention is by building least square method supporting vector machine LS-SVM parameter identification device, utilize Nonlinear Mapping, Hilbert space by input vector from former spatial mappings to higher-dimension, adopt the minimum concept of loss function, utilize the SVM kernel function in former space to replace the inner product operation of high-dimensional feature space, thereby nonlinear function estimation problem is converted into the linear function problem in high-dimensional feature space, be the random deviation dynamic model that concept converts system to by multiple nonlinear random statistical data by SVM kernel function, N-1 impact converted in the kernel function of measured variable by the nonlinear characteristic of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, identification precision is high, its forecast model is the MA model of layering, and the output z (k) of forecast model is its average z ₀and random deviation dynamic model output z (k) _e(k) synthetic, little on affecting the predicated error of key variables of Circulating Fluidized Bed Boiler stable operation, forecasting efficiency is high, time saving and energy saving.

3, application value of the present invention is high, can improve boiler combustion efficiency, lay the foundation for turbine generator stable operation.

4, the present invention can also be applied to parameter identification, prediction and the fault diagnosis of other complex nonlinear multivariable process control, for accurately controlling and lay a good foundation.

Below by drawings and Examples, technical scheme of the present invention is described in further detail.

Accompanying drawing explanation

Fig. 1 is the schematic block circuit diagram of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter identification system of the present invention.

Fig. 2 is the method flow diagram of identification of the present invention and Forecasting Methodology.

Fig. 3 is the measured value of coal for circulation fluid bed boiler spoil flow of the present invention and the comparison diagram of real-time estimate value.

Fig. 4 is the curve map of the real-time estimate value of coal for circulation fluid bed boiler spoil flow of the present invention and the error e 1 of measured value.

Fig. 5 is the measured value of Circulating Fluidized Bed Boiler returning charge blower fan electric current of the present invention and the comparison diagram of real-time estimate value.

Fig. 6 is the curve map of the real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower fan electric current of the present invention and the error e 2 of measured value.

Fig. 7 is the measured value of Circulating Fluidized Bed Boiler flue gas oxygen content of the present invention and the comparison diagram of real-time estimate value.

Fig. 8 is the curve map of the real-time estimate value of Circulating Fluidized Bed Boiler flue gas oxygen content of the present invention and the error e 3 of measured value.

Fig. 9 is the measured value of Circulating Fluidized Bed Boiler flue gas oxygen content while adopting improved BP neural network prediction and the comparison diagram of predicted value.

The curve map of the predicted value of Circulating Fluidized Bed Boiler flue gas oxygen content and the error e 4 of measured value when Figure 10 is the improved BP neural network prediction of employing.

Description of reference numerals:

1-gangue flow sensor; 2-induced draft fan current transformer;

3-returning charge blower fan current transformer; 4-returning charge blower pressure sensor;

5-primary air fan current transformer; 6-overfire air fan current transformer;

7-flue gas oxygen content sensor; 8-data collecting card; 9-processor;

10-display.

Embodiment

As depicted in figs. 1 and 2, identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter of the present invention, comprise the following steps:

Step 1, on affect burning in circulating fluid bed boiler efficiency multiple factor signals real-time detection and synchronously upload: by multiple sensors, the multiple factor signals that affect burning in circulating fluid bed boiler efficiency are detected in real time, and real-time detected signal are synchronously uploaded to data collecting card 8;

Step 2, signal sampling and pre-service: data collecting card 8 carries out L sampling to the detected signal of multiple sensors and correspondingly amplifies, after filtering and A/D conversion process, obtains L × N dimension matrix samples signal x _kiand be synchronously uploaded to processor 9 and carry out record; Wherein, k=1,2 ..., L, i=1,2 ..., N, the value of N equates with the quantity of sensor and N is greater than 2 natural number, L is greater than 20 natural number; In the present embodiment, in step 2, the value of L is 20～50.

In the present embodiment, the quantity of sensor described in step 1 and step 2 is 7, and 7 described sensors are respectively gangue flow sensor 1, induced draft fan current transformer 2, returning charge blower fan current transformer 3, returning charge blower pressure sensor 4, primary air fan current transformer 5, overfire air fan current transformer 6 and flue gas oxygen content sensor 7.

In the present embodiment, a Circulating Fluidized Bed Boiler is surveyed, to 7 detected signals of sensor every sampling in 10 minutes successively, L × N that sampling obtains ties up matrix samples signal x to data collecting card 8 _kiin partial data as shown in table 1:

Table 1 Circulating Fluidized Bed Boiler actual measurement partial data table

In table 1, x _k1represent the measured value of gangue flow, unit is Th ^-1; x _k2represent the measured value of induced draft fan electric current, unit is A; x _k3represent the measured value of returning charge blower fan electric current, unit is A; x _k4represent the measured value of returning charge blower pressure, unit is kPa; x _k5represent the measured value of primary air fan electric current, unit is A; x _k6represent the measured value of overfire air fan electric current, unit is A; x _k7represent the measured value of flue gas oxygen content, unit is %.

The identification of step 3, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the L obtaining in step 2 × N is tieed up to matrix samples signal x by processor 9 _kicarry out analyzing and processing and identification, draw N parameter of any one tested key variables of Circulating Fluidized Bed Boiler, its analyzing and processing and identification process are as follows:

301, processor 9 is to L × N dimension matrix samples signal x _kicarry out analyzing and processing, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz, its analyzing and processing process is as follows:

3011, select the RBF kernel function of SVM: for L sample signal of any one crucial measured variable of Circulating Fluidized Bed Boiler, i.e. L × N ties up matrix samples signal x _kiany one row column vector x _i=[x _1ix _2ix _li] ^t, wherein, i represents columns and i=1～N, the RBF kernel function of its SVM is:

k(x _t,x _j)=exp(-||x _t-x _j|| ²/2s ²) （1-1）

Wherein, x _trepresent that any one crucial measured variable of Circulating Fluidized Bed Boiler is at the measured value in t moment, x _tvalue be column vector x _iin any one value; x _jrepresent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler, x _jvalue be column vector x _ithe average of middle all values; S represents the number of getting SVM;

Known to the analysis of SVM kernel function through experiment, in the time of s=10, test target sample identification rate is 0.862; In the time of s=50, test target sample identification rate is 0.906.Therefore,, in the present embodiment, the value of s is 10～50 in 3011.

3012, processor 9 is according to the RBF kernel function of selecting in 3011, by L × N dimension matrix samples signal x _kifrom N dimension real domain space R ⁿbe mapped to higher-dimension Hilbert space, in conjunction with batch processing least square method, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz:

$H_{\mathrm{Lz}}={\left[\begin{array}{ccccc}{k}_z(x_{1z},x_{\mathrm{zj}})& k_2(x_{12},x_{2j})& k_3(x_{13},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_z(x_{1N},x_{\mathrm{Nj}})\\ k_z(x_{2z},x_{\mathrm{zj}})& k_2(x_{22},x_{2j})& k_3(x_{23},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{2N},x_{\mathrm{Nj}})\\ k_z(x_{3z},x_{\mathrm{zj}})& k_2(x_{32},x_{2j})& k_3(x_{33},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{3N},x_{\mathrm{Nj}})\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_z({\mathrm{<figure-callout\; id="2"\; label="x\; L"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_L,x_{\mathrm{zj}})& k_2({\mathrm{<figure-callout\; id="2"\; label="x\; L"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_L,x_{2j})& k_3({\mathrm{<figure-callout\; id="3"\; label="x\; L"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_L,x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{\mathrm{LN}},x_{\mathrm{Nj}})\end{array}\right]}^T---(1-2)$

Wherein, multidimensional deviation sample kernel function collection H _lzthe first row row vector k _z(x _1z, x _zj), k _z(x _2z, x _zj), k _z(x _3z, x _zj) ..., k _z(x _lz, x _zj) be L sample kernel function of crucial measured variable, multidimensional deviation sample kernel function collection H _lzi every trade vector k _i(x _1i, x _ij), k _i(x _2i, x _ij), k _i(x _ki, x _ij) ..., k _i(x _li, x _ij) be L sample kernel function of the relevant measured variable of i the crucial measured variable of impact, i=2～N;

For example, get N=7, the crucial measured variable of Circulating Fluidized Bed Boiler is gangue flow, the relevant measured variable that affects gangue flow has 6, the relevant measured variable of second crucial measured variable of impact is induced draft fan electric current, the relevant measured variable of the 3rd the crucial measured variable of impact is returning charge blower fan electric current, the relevant measured variable of the 4th the crucial measured variable of impact is returning charge blower pressure, the relevant measured variable of the 5th the crucial measured variable of impact is primary air fan electric current, the relevant measured variable of the 6th the crucial measured variable of impact is overfire air fan electric current, the relevant measured variable of the 7th the crucial measured variable of impact is flue gas oxygen content, multidimensional deviation sample kernel function collection H _lzfirst row column vector k _z(x _1z, x _zj), k _z(x _2z, x _zj), k _z(x _3z, x _zj) ..., k _z(x _lz, x _zj) be L sample kernel function of gangue flow, data collecting card 8 is x to the detected signal of gangue flow sensor 1 measured value obtaining of sampling for the first time _1z, data collecting card 8 is x to the detected signal of gangue flow sensor 1 measured value obtaining of sampling for the second time _2z, data collecting card 8 is x to the detected signal of gangue flow sensor 1 measured value obtaining of sampling for the third time _3z, it is x that data collecting card 8 carries out the L time measured value of obtaining of sampling to the detected signal of gangue flow sensor 1 _lz, the average that data collecting card 8 carries out L all measured values of obtaining of sampling to the detected signal of gangue flow sensor 1 is x _zj, multidimensional deviation sample kernel function collection H _lzsecondary series column vector k ₂(x ₁₂, x _2j), k ₂(x ₂₂, x _2j), k ₂(x ₃₂, x _2j) ..., k ₂(x _l2, x _2j) be L sample kernel function of induced draft fan electric current, data collecting card 8 is x to the detected signal of induced draft fan current transformer 2 measured value obtaining of sampling for the first time ₁₂, data collecting card 8 is x to the detected signal of induced draft fan current transformer 2 measured value obtaining of sampling for the second time ₂₂, data collecting card 8 is x to the detected signal of induced draft fan current transformer 2 measured value obtaining of sampling for the third time ₃₂, it is x that data collecting card 8 carries out the L time measured value of obtaining of sampling to the detected signal of induced draft fan current transformer 2 _l2, the average that data collecting card 8 carries out L all measured values of obtaining of sampling to the detected signal of induced draft fan current transformer 2 is x _2j, by that analogy, multidimensional deviation sample kernel function collection H _lzn(N=7) row column vector k ₇(x ₁₇, x _7j), k ₇(x ₂₇, x _7j), k ₇(x ₃₇, x _7j) ..., k ₇(x _l7, x _7j) be L sample kernel function of flue gas oxygen content, data collecting card 8 is x to the detected signal of flue gas oxygen content sensor 7 measured value obtaining of sampling for the first time ₁₇, data collecting card 8 is x to the detected signal of flue gas oxygen content sensor 7 measured value obtaining of sampling for the second time ₂₇, data collecting card 8 is x to the detected signal of flue gas oxygen content sensor 7 measured value obtaining of sampling for the third time ₃₇, it is x that data collecting card 8 carries out the L time measured value of obtaining of sampling to the detected signal of flue gas oxygen content sensor 7 _l7, the average that data collecting card 8 carries out L all measured values of obtaining of sampling to the detected signal of flue gas oxygen content sensor 7 is x _7j,

In the present embodiment, according to the front 30 groups of data in the measured data in step 2 table 1, obtain 7 of coal for circulation fluid bed boiler spoil flow by the multidimensional deviation sample kernel function collection H of identified parameters _lzas follows:

H _Lz=[2.32.73.12.41.60.60.71.10.3-0.3-0.5-0.9-1.3-0.51.11.6-0.3-3.5-1.7-0.91.60.60.71.10.3-0.3-0.5-0.9-1.3-0.5;

-0.3-0.32-0.48-0.39-0.28-0.08-0.20-0.320.00.20.30.40.480.38-0.020.20.11.00.720.4-0.28-0.08-0.20-0.320.00.20.30.40.480.38;

1.01.32.00.60.80.20.40.70.30.20.1-1.2-1.8-1.0-1.4-1.7-1.9-2.0-1.8-1.60.80.20.40.70.30.20.1-1.2-1.8-1.0;

2.82.43.23.02.61.82.02.30.60.4-1.2-1.4-1.6-1.22.72.80.0-3.4-2.0-1.42.61.82.02.30.60.4-1.2-1.4-1.6-1.2;

3.13.23.83.12.61.01.21.30.0-0.7-1.6-2.2-2.7-1.62.12.61.0-4.0-2.9-2.22.61.01.21.30.0-0.7-1.6-2.2-2.7-1.6;

11.41.81.71.10.20.30.80.1-0.8-0.9-1.0-1.6-0.90.80.60.4-2-1.5-1.01.10.20.30.80.1-0.8-0.9-1.0-1.6-0.9;

1.11.81.91.80.80.10.20.60.4-0.1-0.2-0.4-0.9-0.20.81.60.1-2.2-0.8-0.40.80.10.20.60.4-0.1-0.2-0.4-0.9-0.2]

302, processor 9 is according to formula

$H_{\mathrm{hz}}={\left(H_{\mathrm{Lz}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H}_{\mathrm{Lz}}^T=\left[\begin{array}{ccccc}{k}_{11}& k_{12}& k_{13}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{1L}\\ k_{21}& k_{22}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{2L}\\ k_{31}& k_{32}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{3L}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_{N1}& k_{N2}& {\mathrm{<figure-callout\; id="3"\; label="k\; N"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">k</figure-callout>}}_N& \&CenterDot;\&CenterDot;\&CenterDot;& k_{\mathrm{NL}}\end{array}\right]---(1-3)$

By multidimensional deviation sample kernel function collection H _lzbe transformed to synthetic nucleus function sample matrix H _hz;

In the present embodiment, according to the H in step 3012 _lzthe synthetic nucleus function sample matrix H that data obtain _hzas follows:

303, processor 9 builds the random deviation dynamic model of any one crucial measured variable of Circulating Fluidized Bed Boiler in higher-dimension Hilbert space:

$z_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})---(1-4)$

And draw random deviation dynamic model parameter set according to batch processing least square method and the identification of least disadvantage function principle $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T,$ Wherein, a representative is by the parameter of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, b _irepresent that i impact is by the parameter of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, represent the identifier of a,

represent b _iidentifier, k _i(x _ti, x _ij) represent that i impact is by the kernel function of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, k _i(x _ti, x _ij) value be multidimensional deviation sample kernel function collection H _lzi every trade vector in any one value; Its identification draws random deviation dynamic model parameter set

process as follows:

3031, processor 9 builds the identifier prediction type of any one crucial measured variable of Circulating Fluidized Bed Boiler according to batch processing least square method:

x _kz(t+1)=h ^Tθ=ax _kz(t)+b ₂x _k2(t)+b ₃x _k3(t)+…+b _Nx _kN(t) （1-5）

Wherein, x _kz(t) represent that any one crucial measured variable of Circulating Fluidized Bed Boiler is in the deviate in t moment, any one crucial measured variable of Circulating Fluidized Bed Boiler is in the difference of the measured value in t moment and all measured value averages of any one crucial measured variable of Circulating Fluidized Bed Boiler; h ^t=[x _kz(t) x _k2(t) x _k3(t) ... x _kN(t)] be L × N dimension matrix samples signal x _kii every trade vector, θ represents the random deviation dynamic model parameter set that identification draws true value, and θ=[a b ₂b ₃b _n] ^t;

In the present embodiment, due to N=7, therefore, the identifier prediction type of any one crucial measured variable of Circulating Fluidized Bed Boiler is:

x _kz(t+1)=h ^Tθ=ax _kz(t)+b ₂x _k2(t)+b ₃x _k3(t)+b ₄x _k4(t)+b ₅x _k5(t)+b ₆x _k6(t)+b ₇x _k7(t)

（1-51）

That is:

$x_{\mathrm{kz}}(t+1)=h^T\mathrm{\&theta;}=\left[\begin{array}{ccccccc}{x}_{\mathrm{kz}}\left(t\right)& x_{k2}\left(t\right)& x_{k3}\left(t\right)& x_{k4}\left(t\right)& x_{k5}\left(t\right)& x_{k6}\left(t\right)& x_{k7}\left(t\right)\end{array}\right]\left[\begin{array}{c}a\\ b_2\\ b_3\\ b_4\\ b_5\\ b_6\\ b_7\end{array}\right]---(1-52)$

In fact, formula (1-52) is x _kz(t+1) least square form, wherein, 7 sensors, the one group of sample recording of once sampling is: h ^t=[x _kz(t) x _k2(t) x _k3(t) x _k4(t) x _k5(t) x _k6(t) x _k7(t)], 7 corresponding parameters are write as column vector and are: θ=[a b ₂b ₃b ₄b ₅b ₆b ₇] ^t;

3032, processor 9 defines the measured value z (k) and its identifier x of any one crucial measured variable that the loss function of identification is Circulating Fluidized Bed Boiler _kz(t+1) Quadratic Function Optimization of the error between, is expressed as:

$J\left(\mathrm{\&theta;}\right)=\frac{1}{2}\underset{k=1}{\overset{L}{\mathrm{\&Sigma;}}}{[z\left(k\right)-h^T\mathrm{\&theta;}]}^2---(1-6)$

Make loss function minimum, meet:

$\left\{\begin{array}{c}\frac{\&PartialD;J\left(\mathrm{\&theta;}\right)}{\&PartialD;\mathrm{\&theta;}}=0\\ \frac{{\&PartialD;}^{2}J\left(\mathrm{\&theta;}\right)}{{\&PartialD;\mathrm{\&theta;}}^2}>0\end{array}\right.---(1-7)$

Obtain the identifier of θ:

$\widehat{\mathrm{\&theta;}}={\left({H_{\mathrm{Lz}}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H_{\mathrm{Lz}}}^T{z}_L=H_{\mathrm{hz}}{z}_L---(1-8)$

Wherein, z _lrepresent in multiple described sensors any one L sampling output valve, the i.e. L of any one crucial measured variable of a Circulating Fluidized Bed Boiler measured value;

Particularly, processor 9 is to formula (1-6) with (1-7) to process the process of the formula of obtaining (1-8) as follows:

$\begin{array}{c}\frac{\&PartialD;J\left(\mathrm{\&theta;}\right)}{\&PartialD;\mathrm{\&theta;}}=\frac{\&PartialD;}{\&PartialD;\mathrm{\&theta;}}{[z_L-H_L\mathrm{\&theta;}]}^T\text{[}{z}_L\text{-}{H}_L\text{\&theta;]=-}{H}_L^T(z_L-H_L\mathrm{\&theta;})-{(z_L-H_L\mathrm{\&theta;})}^T{H}_L\\ =-H_L^T(z_L-H_L\mathrm{\&theta;})-H_L^T(z_L-H_L\mathrm{\&theta;})=-2H_L^T(z_L-H_L\mathrm{\&theta;})=-2H_L^T{z}_L+{2H}_L^T{H}_L\mathrm{\&theta;}=0\end{array}---(1-71)$

$\frac{{\&PartialD;}^{2}J\left(\mathrm{\&theta;}\right)}{{\&PartialD;\mathrm{\&theta;}}^2}={2H}_L^T{H}_L>0---(1-72)$

Drawn by formula (1-71):

$H_L^T{z}_L=H_L^T{H}_L\mathrm{\&theta;}---(1-73)$

Obtained the estimated value of θ by formula (1-73) $\widehat{\mathrm{\&theta;}}={\left({H_{\mathrm{Lz}}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H_{L2}}^T{z}_L=H_{\mathrm{hz}}{z}_L;$

3033, processor 9 is according to formula

$\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T=H_{\mathrm{hz}}{z}_L---(1-9)$

Identification draws the random deviation dynamic model parameter set of any one crucial measured variable of Circulating Fluidized Bed Boiler

be N parameter of identification any one key variables of having drawn Circulating Fluidized Bed Boiler;

In the present embodiment, the random deviation dynamic model parameter set of the coal for circulation fluid bed boiler spoil flow that identification draws $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& {\hat{b}}_4& {\hat{b}}_5& {\hat{b}}_6& {\hat{b}}_7\end{array}\right]}^T=\left[\begin{array}{ccc}1.0036& 0.1532& 0.0127\end{array}\right.$ ${\left.\begin{array}{cccc}0.0065& -0.0034& 0.0525& -0.0366\end{array}\right]}^T,$ 7 parameters that are coal for circulation fluid bed boiler spoil flow are respectively: $\hat{a}=1.0036,{\hat{b}}_2=0.1532,{\hat{b}}_3=0.0137,{\hat{b}}_4=0.0065,{\hat{b}}_5=-0.0034,$ ${\hat{b}}_6=0.0525,{\hat{b}}_7=-0.0366.$

The real-time estimate of step 4, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the random deviation dynamic model of any one crucial measured variable of the Circulating Fluidized Bed Boiler of processor (9) based on building in 303, builds the hierarchical prediction model of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter:

$\left\{\begin{array}{c}{z}_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})\\ z(k+1)=z_0\left(k\right)+z_e\left(k\right)\end{array}\right.---(1-10)$

And according to the random deviation dynamic model parameter set of any one crucial measured variable of the Circulating Fluidized Bed Boiler that in hierarchical prediction model and 303, identification draws

dope the real-time estimate value z (k+1) of any one crucial measured variable of Circulating Fluidized Bed Boiler; Wherein, z ₀(k) represent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler;

Particularly, the ground floor in hierarchical prediction model is the random deviation dynamic model z of any one crucial measured variable of the Circulating Fluidized Bed Boiler building in 303 _e(k), the second layer in hierarchical prediction model is by the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler and random deviation dynamic model z _e(k) synthesize, obtained the real-time estimate value z (k+1) of any one crucial measured variable of Circulating Fluidized Bed Boiler;

In the present embodiment, the hierarchical prediction model of the coal for circulation fluid bed boiler spoil flow of structure is as follows:

$\left\{\begin{array}{c}{z}_e\left(k\right)=1.0036+0.1532k_2({\mathrm{<figure-callout\; id="2"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})+0.0137k_3({\mathrm{<figure-callout\; id="3"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})+0.0065k_4({\mathrm{<figure-callout\; id="4"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{4j})\\ -0.0034k_5({\mathrm{<figure-callout\; id="5"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE001.png,2012101462899100003DEST\_PATH\_IMAGE003.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{5j})+0.0525k_6(x_{t6},x_{6j})-0.0366k_7({\mathrm{<figure-callout\; id="7"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{7j})\\ z(k+1)=38.5+z_e\left(k\right)\end{array}\right.---(1-101)$

In formula (1-101), z (k+1) is the real-time estimate value of coal for circulation fluid bed boiler spoil flow, the average of 38.5 all measured values that are the coal for circulation fluid bed boiler spoil flow that obtains according to the measured data in table 1.

Get in step 2 table 1 in measured data front 30 groups of measured values of coal for circulation fluid bed boiler spoil flow as follows:

30 groups of real-time estimate values predicting the coal for circulation fluid bed boiler spoil flow obtaining according to formula (1-101) are as follows:

The real-time estimate value of coal for circulation fluid bed boiler spoil flow and the error e of measured value 1 are:

Draw out 30 groups of measured values of above coal for circulation fluid bed boiler spoil flow and the comparison diagram of 30 groups of real-time estimate values as shown in Figure 3 by processor 9, in Fig. 3, Ims represents the measured value of coal for circulation fluid bed boiler spoil flow, and Imy represents the real-time estimate value of coal for circulation fluid bed boiler spoil flow; The curve map of drawing out the real-time estimate value of coal for circulation fluid bed boiler spoil flow and the error e 1 of measured value by processor 9 as shown in Figure 4.

Can find out from data, Fig. 3 and Fig. 4 of error e 1, the real-time estimate value of coal for circulation fluid bed boiler spoil flow and the error e 1 of measured value are less than 0.6%.

Step 5, repeat step 3 and step 4, until identification draws all parameters of all crucial measured variables of Circulating Fluidized Bed Boiler, and dope the real-time estimate value of all crucial measured variables of Circulating Fluidized Bed Boiler; Can improve the burning efficiency of Circulating Fluidized Bed Boiler, for turbine generator stable operation lays the foundation.

In the present embodiment, all crucial measured variable of Circulating Fluidized Bed Boiler described in step 5 is gangue flow, returning charge blower pressure and three crucial measured variables of flue gas oxygen content; Each crucial measured variable all has seven identified parameters.

In the present embodiment, the random deviation dynamic model parameter set of the Circulating Fluidized Bed Boiler returning charge blower pressure that identification draws $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& {\hat{b}}_4& {\hat{b}}_5& {\hat{b}}_6& {\hat{b}}_7\end{array}\right]}^T=\left[\begin{array}{ccc}1.0149& -0.0378& -0.0266\end{array}\right.$ ${\left.\begin{array}{cccc}0.0097& -0.0069& -0.0439& 0.0417\end{array}\right]}^T,$ 7 parameters that are coal for circulation fluid bed boiler spoil flow are respectively: $\hat{a}=1.0149,{\hat{b}}_2=-0.0378,{\hat{b}}_3=-0.0266,{\hat{b}}_4=0.0097,{\hat{b}}_5=0.0069,$ ${\hat{b}}_6=-0.0439,{\hat{b}}_7=0.0417.$

In the present embodiment, the hierarchical prediction model of the Circulating Fluidized Bed Boiler returning charge blower pressure of structure is as follows:

$\left\{\begin{array}{c}{z}_e\left(k\right)=1.0149-0.0378k_2({\mathrm{<figure-callout\; id="2"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})-0.0266k_3({\mathrm{<figure-callout\; id="3"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})+0.0097k_4({\mathrm{<figure-callout\; id="4"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{4j})\\ +0.0069k_5({\mathrm{<figure-callout\; id="5"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE001.png,2012101462899100003DEST\_PATH\_IMAGE003.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{5j})-0.0439k_6(x_{t6},x_{6j})+0.0417k_7({\mathrm{<figure-callout\; id="7"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{7j})\\ z(k+1)=44.0+z_e\left(k\right)\end{array}\right.---(1-102)$

In formula (1-102), z (k+1) is the real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower pressure, the average of 44.0 all measured values that are the Circulating Fluidized Bed Boiler returning charge blower pressure that obtains according to the measured data in table 1.

Get in step 2 table 1 in measured data front 30 groups of measured values of Circulating Fluidized Bed Boiler returning charge blower pressure as follows:

30 groups of real-time estimate values predicting the Circulating Fluidized Bed Boiler returning charge blower pressure obtaining according to step 4 are as follows:

The real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower pressure and the error e of measured value 2 are:

Draw out 30 groups of measured values of Circulating Fluidized Bed Boiler returning charge blower pressure and the comparison diagram of 30 groups of real-time estimate values as shown in Figure 5 by processor 9, in Fig. 5, Ps represents the measured value of Circulating Fluidized Bed Boiler returning charge blower pressure, and Py represents the real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower pressure; The curve map of drawing out the real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower pressure and the error e 2 of measured value by processor 9 as shown in Figure 6.

Can find out from Fig. 5 and Fig. 6, the real-time estimate value of Circulating Fluidized Bed Boiler returning charge blower pressure and the error e 2 of measured value are less than 0.07%.

In the present embodiment, the random deviation dynamic model parameter set of the Circulating Fluidized Bed Boiler flue gas oxygen content that identification draws $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& {\hat{b}}_4& {\hat{b}}_5& {\hat{b}}_6& {\hat{b}}_7\end{array}\right]}^T=\left[\begin{array}{cccc}1.0572& 0.0049& 0.0080& 0.0004\end{array}\right.$ ${\left.\begin{array}{ccc}-0.0117& 0.0272& 0.0066\end{array}\right]}^{\mathrm{<figure-callout\; id="7"\; label="T"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">T</figure-callout>}}$ 7 parameters that are Circulating Fluidized Bed Boiler flue gas oxygen content are respectively: $\hat{a}=1.0572,{\hat{b}}_2=0.0049,{\hat{b}}_3=0.0080,{\hat{b}}_4=0.0004,{\hat{b}}_5=-0.0117,{\hat{b}}_6=0.0272,{\hat{b}}_7=0.0066.$

In the present embodiment, the hierarchical prediction model of the Circulating Fluidized Bed Boiler flue gas oxygen content of structure is as follows:

$\left\{\begin{array}{c}{z}_e\left(k\right)=1.0572+0.0049k_2({\mathrm{<figure-callout\; id="2"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})+0.0080k_3({\mathrm{<figure-callout\; id="3"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png,HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{2j})+0.0004k_4({\mathrm{<figure-callout\; id="4"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE007.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{4j})\\ -0.0117k_5({\mathrm{<figure-callout\; id="5"\; label="x\; t"\; filenames="2012101462899100003DEST\_PATH\_IMAGE001.png,2012101462899100003DEST\_PATH\_IMAGE003.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{5j})+0.0272k_6(x_{t6},x_{6j})+0.0066k_7({\mathrm{<figure-callout\; id="7"\; label="x\; t"\; filenames="HDA00001630171400011.png"\; state="\{\{state\}\}">x</figure-callout>}}_t,x_{7j})\\ z(k+1)=4.5+z_e\left(k\right)\end{array}\right.---(1-103)$

In formula (1-103), z (k+1) is the real-time estimate value of Circulating Fluidized Bed Boiler flue gas oxygen content, the average of 4.0 all measured values that are the Circulating Fluidized Bed Boiler flue gas oxygen content that obtains according to the measured data in table 1.

Get in step 2 table 1 in measured data front 30 groups of measured values of Circulating Fluidized Bed Boiler flue gas oxygen content as follows:

30 groups of real-time estimate values predicting the Circulating Fluidized Bed Boiler flue gas oxygen content obtaining according to step 4 are as follows:

The real-time estimate value of recirculating fluidized bed pot flue gas oxygen content and the error e of measured value 3 are:

Draw out 30 groups of measured values of Circulating Fluidized Bed Boiler flue gas oxygen content and the comparison diagram of 30 groups of real-time estimate values as shown in Figure 7 by processor 9, in Fig. 7, O2s represents the measured value of Circulating Fluidized Bed Boiler flue gas oxygen content, and O2y represents the real-time estimate value of Circulating Fluidized Bed Boiler flue gas oxygen content; The curve map of drawing out the real-time estimate value of Circulating Fluidized Bed Boiler flue gas oxygen content and the error e 3 of measured value by processor 9 as shown in Figure 8.

As can be seen from Figures 7 and 8, the real-time estimate value of Circulating Fluidized Bed Boiler flue gas oxygen content and the error e 3 of measured value are less than 0.7%.

Step 6, result are synchronously exported: in step 3, carry out in the identification process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and in step 4, carry out in the real-time estimate process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, processor 9 by the display 10 that joins with it in the identification result to the signal processing in step 3 and Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and step 4 predicting the outcome of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter carry out simultaneous display.

In the present embodiment, described processor 9 is industrial control computer.While carrying out the identification of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 3, and while carrying out the real-time estimate of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 4, described industrial control computer is realized by MATLAB, Visual C and configuration software.While building the random deviation dynamic model of any one crucial measured variable of Circulating Fluidized Bed Boiler in 303, be selected slip model (be called for short MA model), then the N of any one crucial measured variable of Circulating Fluidized Bed Boiler be updated to the random deviation dynamic model that obtains any one crucial measured variable of the represented Circulating Fluidized Bed Boiler of formula (1-4) in slip model by identified parameters.

In order further identification and the precision of prediction of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter of the present invention to be verified, according to 30 groups of measured values of same Circulating Fluidized Bed Boiler flue gas oxygen content, and adopt " 7-21-1 " type (to input node i=7, hidden node j=21, output node k=1) improved BP neural network predicts Circulating Fluidized Bed Boiler flue gas oxygen content:

The initial value in neural network hidden layer weights (k-1) moment:

The initial value in neural network hidden layer weights (k-2) moment:

The initial value in neural network output weights (k-1) moment:

The initial value in neural network output weights (k-2) moment:

The predicted value of neural network to flue gas oxygen content:

Neural network prediction result hidden layer weights

Neural network prediction result output layer weights:

Draw out 30 groups of measured values of Circulating Fluidized Bed Boiler flue gas oxygen content and the comparison diagram of 30 groups of predicted values as shown in Figure 9 by processor 9, in Fig. 9, O2s represents the measured value of Circulating Fluidized Bed Boiler flue gas oxygen content, and O2y of NN representative adopts the predicted value of the Circulating Fluidized Bed Boiler flue gas oxygen content of the improved BP neural network of " 7-21-1 " type; Draw out the predicted value of Circulating Fluidized Bed Boiler flue gas oxygen content of the improved BP neural network of employing " 7-21-1 " type and the curve map of the error e of measured value 4 as shown in figure 10 by processor 9.

Can find out from Fig. 9 and Figure 10, adopt the predicted value of Circulating Fluidized Bed Boiler flue gas oxygen content and the error e of measured value 4 of the improved BP neural network of " 7-21-1 " type to be 2.3% to the maximum.

From contrasting, identification of the present invention and precision of prediction are far away higher than the precision of prediction of the improved BP neural network of " 7-21-1 " type.

In order further identification and the precision of prediction of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter of the present invention to be verified again, get again from above-mentioned identification and when prediction measured value differ one group of data that 30%-50% is different and verify, the predicated error of gangue flow, returning charge blower pressure and flue gas oxygen content is all less than to 0.7%.

The present invention can also be applied to parameter identification, prediction and the fault diagnosis of other complex nonlinear multivariable process control, for accurately controlling and lay a good foundation.

The above; it is only preferred embodiment of the present invention; not the present invention is imposed any restrictions, every any simple modification of above embodiment being done according to the technology of the present invention essence, change and equivalent structure change, and all still belong in the protection domain of technical solution of the present invention.

Claims (6)

1. the identification of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and a Forecasting Methodology, is characterized in that the method comprises the following steps:

Step 1, on affect burning in circulating fluid bed boiler efficiency multiple factor signals real-time detection and synchronously upload: by multiple sensors, the multiple factor signals that affect burning in circulating fluid bed boiler efficiency are detected in real time, and real-time detected signal are synchronously uploaded to data collecting card (8);

Step 2, signal sampling and pre-service: data collecting card (8) carries out L sampling to the detected signal of multiple sensors and correspondingly amplifies, after filtering and A/D conversion process, obtains L × N dimension matrix samples signal x _kiand be synchronously uploaded to processor (9) and carry out record; Wherein, k=1,2 ..., L, i=1,2 ..., N, the value of N equates with the quantity of sensor and N is greater than 2 natural number, L is greater than 20 natural number;

The identification of step 3, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the L obtaining in step 2 × N is tieed up to matrix samples signal x by processor (9) _kicarry out analyzing and processing and identification, draw N parameter of any one tested key variables of Circulating Fluidized Bed Boiler, its analyzing and processing and identification process are as follows:

301, processor (9) is to L × N dimension matrix samples signal x _kicarry out analyzing and processing, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz, its analyzing and processing process is as follows:

3011, select the RBF kernel function of SVM: for L sample signal of any one crucial measured variable of Circulating Fluidized Bed Boiler, i.e. L × N ties up matrix samples signal x _kiany one row column vector x _i=[x _1ix _2ix _li] ^t, wherein, i represents columns and i=1～N, the RBF kernel function of its SVM is:

k(x _t,x _j)=exp(-||x _t-x _j|| ²/2s ²) （1-1）

Wherein, x _trepresent that any one crucial measured variable of Circulating Fluidized Bed Boiler is at the measured value in t moment, x _tvalue be column vector x _iin any one value; x _jrepresent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler, x _jvalue be column vector x _ithe average of middle all values; S represents the number of getting SVM;

3012, processor (9) is according to the RBF kernel function of selecting in 3011, by L × N dimension matrix samples signal x _kifrom N dimension real domain space R ⁿbe mapped to higher-dimension Hilbert space, in conjunction with batch processing least square method, show that the N of any one crucial measured variable of Circulating Fluidized Bed Boiler is by the multidimensional deviation sample kernel function collection H of identified parameters _lz:

$H_{\mathrm{Lz}}={\left[\begin{array}{ccccc}{k}_z(x_{1z},x_{\mathrm{zj}})& k_2(x_{12},x_{2j})& k_3(x_{13},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_z(x_{1N},x_{\mathrm{Nj}})\\ k_z(x_{2z},x_{\mathrm{zj}})& k_2(x_{22},x_{2j})& k_3(x_{23},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{2N},x_{\mathrm{Nj}})\\ k_z(x_{3z},x_{\mathrm{zj}})& k_2(x_{32},x_{2j})& k_3(x_{33},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{3N},x_{\mathrm{Nj}})\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_z(x_{L2},x_{\mathrm{zj}})& k_2(x_{L2},x_{2j})& k_3(x_{L3},x_{3j})& \&CenterDot;\&CenterDot;\&CenterDot;& k_N(x_{\mathrm{LN}},x_{\mathrm{Nj}})\end{array}\right]}^T---(1-2)$

Wherein, multidimensional deviation sample kernel function collection H _lzthe first row row vector k _z(x _1z, x _zj), k _z(x _2z, x _zj), k _z(x _3z, x _zj) ..., k _z(x _lz, x _zj) be L sample kernel function of crucial measured variable, multidimensional deviation sample kernel function collection H _lzi every trade vector k _i(x _1i, x _ij), k _i(x _2i, x _ij), k _i(x _ki, x _ij) ..., k _i(x _li, x _ij) be L sample kernel function of the relevant measured variable of i the crucial measured variable of impact, i=2～N;

302, processor (9) is according to formula

$H_{\mathrm{hz}}={\left(H_{\mathrm{Lz}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H}_{\mathrm{Lz}}^T=\left[\begin{array}{ccccc}{k}_{11}& k_{12}& k_{13}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{1L}\\ k_{21}& k_{22}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{2L}\\ k_{31}& k_{32}& k_{33}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{3L}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ k_{N1}& k_{N2}& k_{N3}& \&CenterDot;\&CenterDot;\&CenterDot;& k_{\mathrm{NL}}\end{array}\right]---(1-3)$

By multidimensional deviation sample kernel function collection H _lzbe transformed to synthetic nucleus function sample matrix H _hz;

303, processor (9) builds the random deviation dynamic model of any one crucial measured variable of Circulating Fluidized Bed Boiler in higher-dimension Hilbert space:

$z_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})---(1-4)$

And draw random deviation dynamic model parameter set according to batch processing least square method and the identification of least disadvantage function principle $\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T,$ Wherein, a representative is by the parameter of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, b _irepresent that i impact is by the parameter of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification,

represent the identifier of a,

represent b _iidentifier, k _i(x _ti, x _ij) represent that i impact is by the kernel function of the relevant measured variable of the crucial measured variable of the Circulating Fluidized Bed Boiler of identification, k _i(x _ti, x _ij) value be multidimensional deviation sample kernel function collection H _lzi every trade vector in any one value; Its identification draws random deviation dynamic model parameter set

process as follows:

3031, processor (9) builds the identifier prediction type of any one crucial measured variable of Circulating Fluidized Bed Boiler according to batch processing least square method:

x _kz(t+1)=h ^Tθ=ax _kz(t)+b ₂x _k2(t)+b ₃x _k3(t)+…+b _Nx _kN(t) （1-5）

Wherein, x _kz(t) represent that any one crucial measured variable of Circulating Fluidized Bed Boiler is in the deviate in t moment, any one crucial measured variable of Circulating Fluidized Bed Boiler is in the difference of the measured value in t moment and all measured value averages of any one crucial measured variable of Circulating Fluidized Bed Boiler; h ^t=[x _kz(t) x _k2(t) x _k3(t) ... x _kN(t)] be L × N dimension matrix samples signal x _kii every trade vector, θ represents the random deviation dynamic model parameter set that identification draws

true value, and θ=[a b ₂b ₃b _n] ^t;

The measured value z (k) and its identifier x of any one crucial measured variable that 3032, the loss function of processor (9) definition identification is Circulating Fluidized Bed Boiler _kz(t+1) Quadratic Function Optimization of the error between, is expressed as:

$J\left(\mathrm{\&theta;}\right)=\frac{1}{2}\underset{k=1}{\overset{L}{\mathrm{\&Sigma;}}}{[z\left(k\right)-h^T\mathrm{\&theta;}]}^2---(1-6)$

Make loss function minimum, meet:

$\left\{\begin{array}{c}\frac{\&PartialD;J\left(\mathrm{\&theta;}\right)}{\&PartialD;\mathrm{\&theta;}}=0\\ \frac{{\&PartialD;}^{2}J\left(\mathrm{\&theta;}\right)}{{\&PartialD;\mathrm{\&theta;}}^2}>0\end{array}\right.---(1-7)$

Obtain the identifier of θ:

$\widehat{\mathrm{\&theta;}}={\left({H_{\mathrm{Lz}}}^T{H}_{\mathrm{Lz}}\right)}^{-1}{H_{\mathrm{Lz}}}^T{z}_L=H_{\mathrm{hz}}{z}_L---(1-8)$

Wherein, z _lrepresent in multiple described sensors any one L sampling output valve, the i.e. L of any one crucial measured variable of a Circulating Fluidized Bed Boiler measured value;

3033, processor (9) is according to formula

$\widehat{\mathrm{\&theta;}}={\left[\begin{array}{ccccc}\hat{a}& {\hat{b}}_2& {\hat{b}}_3& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{b}}_N\end{array}\right]}^T=H_{\mathrm{hz}}{z}_L---(1-9)$

Identification draws the random deviation dynamic model parameter set of any one crucial measured variable of Circulating Fluidized Bed Boiler

be N parameter of identification any one key variables of having drawn Circulating Fluidized Bed Boiler;

The real-time estimate of step 4, Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter: the random deviation dynamic model of any one crucial measured variable of the Circulating Fluidized Bed Boiler of processor (9) based on building in 303, builds the hierarchical prediction model of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter:

$\left\{\begin{array}{c}{z}_e\left(k\right)=a+\underset{i=2}{\overset{N}{\mathrm{\&Sigma;}}}{b}_i{k}_i(x_{\mathrm{ti}},x_{\mathrm{ij}})\\ z(k+1)=z_0\left(k\right)+z_e\left(k\right)\end{array}\right.---(1-10)$

And according to the random deviation dynamic model parameter set of any one crucial measured variable of the Circulating Fluidized Bed Boiler that in hierarchical prediction model and 303, identification draws

dope the real-time estimate value z (k+1) of any one crucial measured variable of Circulating Fluidized Bed Boiler; Wherein, z ₀(k) represent the average of all measured values of any one crucial measured variable of Circulating Fluidized Bed Boiler;

Step 5, repeat step 3 and step 4, until identification draws all parameters of all crucial measured variables of Circulating Fluidized Bed Boiler, and dope the real-time estimate value of all crucial measured variables of Circulating Fluidized Bed Boiler;

Step 6, result are synchronously exported: in step 3, carry out in the identification process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and in step 4, carry out in the real-time estimate process of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter, processor (9) by the display (10) that joins with it in the identification result to the signal processing in step 3 and Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter and step 4 predicting the outcome of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter carry out simultaneous display.

2. according to identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter claimed in claim 1, it is characterized in that: the quantity of sensor described in step 1 and step 2 is 7,7 described sensors are respectively gangue flow sensor (1), induced draft fan current transformer (2), returning charge blower fan current transformer (3), returning charge blower pressure sensor (4), primary air fan current transformer (5), overfire air fan current transformer (6) and flue gas oxygen content sensor (7).

3. according to identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter claimed in claim 1, it is characterized in that: all crucial measured variable of Circulating Fluidized Bed Boiler described in step 5 is gangue flow, returning charge blower pressure and flue gas oxygen content.

4. according to identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter claimed in claim 1, it is characterized in that: described processor (9) is industrial control computer.

5. according to identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter claimed in claim 4, it is characterized in that: while carrying out the identification of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 3, and while carrying out the real-time estimate of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter in step 4, described industrial control computer is realized by MATLAB, Visual C and configuration software.

6. according to identification and the Forecasting Methodology of Circulating Fluidized Bed Boiler Nonlinear Multivariable key parameter claimed in claim 1, it is characterized in that: in step 2 and step 3, the value of L is 20～50; In 3011, the value of s is 10～50.

Images