CN105243178A - System and method for quantifying coal supply heat release time of circulating fluidized bed boiler - Google Patents

Abstract

The invention relates to a system and method for quantifying coal supply heat release time of a circulating fluidized bed boiler. The system comprises a least-square support vector machine modeling module, a data selection and preprocessing module, a coal supply heat release time optimizing module, a DCS system and a database. According to the system, the coal supply heat release time of the circulating fluidized bed boiler in different load sections and different operational conditions is quantified, and the coal supply heat release time in different loads is quantified by establishing a quantification model for the coal supply heat release time of the circulating fluidized bed boiler and fully considering delay, inertial and heat storage difference of a circulating fluidized bed boiler unit in different load sections.

Description

Quantization loop fluidized-bed combustion boiler coal supply thermal release time system and method

Technical field

The present invention relates to energy project Circulating Fluidized Bed Boiler field, particularly relate to a kind of quantization loop fluidized-bed combustion boiler coal supply thermal release time system and method.

Background technology

Circulating Fluidized Bed Boiler has the unique advantage such as direct desulfurization and burning coal inferior in stove, is the emphasis of development clean coal combustion technology, obtains applying more and more widely over nearly 20 years.Along with the continuous lifting of Circulating Fluidized Bed Boilers capacity, accumulation of heat, the inertia of boiler increase severely, after coal-supplying amount enters burner hearth, the dispose procedure of heat differs greatly under different load sections, and be subject to the combined influence of the factors such as other such as first and second air quantity, be difficult to obtain accurate, comprehensive information by the test at scene.Because actual set is difficult to determine the coal supply thermal release time, cause the isoparametric lag fluctuation of bed temperature obvious, bring difficulty to the automatic control of unit.For requirements such as responsive electricity grid side frequency modulation, the load change of boiler is more and more frequent, determine that the coal supply thermal release time is most important, set up quantization system and the method for Coal Feeding in Circulating Fluidized Bed Furnace thermal release time, optimize significant to the safe operation of unit, quick load change and control system.

Summary of the invention

The present invention is directed to the phenomenon that the current Coal Feeding in Circulating Fluidized Bed Furnace thermal release time is difficult to quantification, a kind of quantization loop fluidized-bed combustion boiler coal supply thermal release time system and method are provided, utilize data unit operation, by setting up the quantitative model of Coal Feeding in Circulating Fluidized Bed Furnace thermal release time, take into full account the delay of Circulating Fluidized Bed Boilers under different load section, inertia and accumulation of heat difference, the coal supply thermal release time under quantification different load.

Quantization loop fluidized-bed combustion boiler coal supply thermal release time system of the present invention comprises:

Modeling method of least squares support module;

Data decimation and pretreatment module;

Coal supply thermal release time optimizing module; And

DCS system and database,

Wherein, described DCS system is connected with described pretreatment module with described data decimation with database, described data decimation and pretreatment module and described modeling method of least squares support module are bi-directionally connected, described modeling method of least squares support module and described coal supply thermal release time optimizing model calling, described coal supply thermal release time optimizing module processes the training data selected and predicted data, determine the coal supply thermal release time of corresponding load section, be transferred to described DCS system and database.

Preferably, described modeling method of least squares support module is used for setting up minimum algorithm of support vector machine coal supply thermal release time model, and the output of model is dynamic bed temperature value, determines to input the thermal release time in coal-supplying amount according to the precision of model prediction.

Preferably, the historical data of unit operation is transferred to described data decimation and pretreatment module by described DCS system and database, described data decimation and pretreatment module are according to the training data of historical data Selection Model and predicted data, and the model training data choosing out and predicted data are transferred to described modeling method of least squares support module by described data decimation and pretreatment module.

Preferably, described modeling method of least squares support module is used for setting up LSSVM model, and the model training data received and predicted data are transferred to described coal supply thermal release time optimizing module by described modeling method of least squares support module.

Preferably, described coal supply thermal release time optimizing module processes the training data received and predicted data, determines optimal dynamic rank group, and then determines the coal supply thermal release time and be transferred to described DCS system and database.

Present invention also offers a kind of method carrying out the quantization loop fluidized-bed combustion boiler coal supply thermal release time according to said system, comprise the following steps:

Step S1, utilizes described modeling method of least squares support module construction least square method supporting vector machine model;

Step S2, utilizes described data decimation and pretreatment module, chooses least square method supporting vector machine Algorithm for Training data in modeling method of least squares support module;

Step S3, the least square method supporting vector machine model that described coal supply thermal release time optimizing module is set up according to step S1, the coal supply thermal release time optimal value in the training data determine described step S2 is chosen.

Preferably, described step S1 least square method supporting vector machine model is:

y(k)＝f[y(k-1),......,y(k-m)；x ₀(k),......,x ₀(k-n ₀)；

x ₁(k),......,x ₁(k-n ₁)；x ₂(k),......,x ₂(k-n ₂)]

Build in described least square method supporting vector machine model, adopt Gaussian radial basis function core, namely

K(x,x _i)＝exp(-||x-x _i|| ²/σ ²)

Wherein said x (k) is the input quantity of model, described x ₀(k) ..., x ₀(k-n ₀), x ₁(k) ..., x ₁(k-n ₁) and described x ₂(k) ..., x ₂(k-n ₂) be respectively corresponding sampling instant in Circulating Fluidized Bed Boiler and determine the coal-supplying amount of bed temperature, primary air flow and secondary air flow; Described y (k) exports for current bed temperature; Described y (k-1) ..., y (k-m) represents that history bed temperature exports, described m, n ₀, n ₁, n ₂represent the dynamic order of history bed temperature, the dynamic order of coal-supplying amount, the dynamic order of primary air flow and the dynamic order of secondary air flow respectively, the precision determination coal supply thermal release time according to model prediction is n ₀be multiplied by data sampling time ts; Described σ is a location parameter.

Preferably, build described least square method supporting vector machine model, adopt least square support vector algorithm, described least square method supporting vector machine algorithm and described Gaussian radial basis function core comprise two unknown parameter c and σ, utilize grid data service and cross validation to carry out.Described step S1 comprises following sub-step:

The Candidate Set of sub-step S1.1, setting c and σ be more open grid (c1, σ 1) ..., (cl, σ l) }, carry out cross validation with the node in grid, obtain the grid node corresponding to least error;

Sub-step S1.2, constructs new grid according to the grid node that above-mentioned sub-step S1.1 obtains, and with the node in grid for parameter is tested, obtains the value of optimum c and σ;

Preferably, described step S2 comprises following sub-step:

Sub-step S2.1, determines sampling time ts, unit: second, obtains historical data from described DCS system and database, comprises time point, unit load, coal-supplying amount, primary air flow, bed temperature value;

Sub-step S2.2, according to the historical data that sub-step S2.1 obtains, a point load section sets up described least square method supporting vector machine model, and subregion scope is between 20 ~ 100WM;

Sub-step S2.3, according between above-mentioned sub-step S2.2 region of differentiation, under selecting different load section, a certain amount of history data is as training data.

Preferably, the optimal value of the dynamic rank group in the training data that described step S3 coal supply thermal release time optimizing module utilizes root-mean-square error to choose described sub-step S2.3 is chosen, and coal supply thermal release time optimal value is n ₀× ts; Described root-mean-square error is used for weighing observed reading with the deviation between true value, and it is the square root of the quadratic sum observation frequency n ratio of observed reading and true value deviation, namely

$RMSE=\sqrt{\underset{i=1}{\overset{n}{\Σ}}{({\hat{y}}_i-y_i)}^2/n}$

Preferably, described coal supply thermal release time optimizing module, the method of traversal is all adopted to carry out under each load section of optimizing part, namely the root-mean-square error often organizing the combination of dynamic rank is calculated, choose the minimum combination of root-mean-square error as optimizing result, if root-mean-square error is less than 1 described in optimizing result, by corresponding for optimizing result m, n ₀, n ₁, n ₂be worth the dynamic rank group optimal value as corresponding load section, will n be worth ₀× ts as the coal supply thermal release time of corresponding load section, otherwise chooses more data from described data decimation and pretreatment module, continues to optimize.

Preferably, the scope of described m is 0 ~ 360/ts, described n ₀scope be 0 ~ 900/ts, described n ₁scope be 0 ~ 240/ts, described n ₂scope be 0 ~ 300/ts, be spaced apart 1.

beneficial effect

Quantization loop fluidized-bed combustion boiler coal supply thermal release time system of the present invention and method have quantized the coal supply thermal release time under different load section, under different operating condition, are conducive to the safe operation of unit, quick load change and control system optimization.

Accompanying drawing explanation

Fig. 1 illustrates the structural drawing of quantization loop fluidized-bed combustion boiler coal supply thermal release time system of the present invention.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is explained in further detail.Should be appreciated that specific embodiment described herein only for explaining the present invention, being not intended to limit the present invention.

On the contrary, the present invention is contained any by the substituting of making on marrow of the present invention and scope of defining of claim, amendment, equivalent method and scheme.Further, in order to make the public have a better understanding to the present invention, in hereafter details of the present invention being described, detailedly describe some specific detail sections.Do not have the description of these detail sections can understand the present invention completely for a person skilled in the art yet.

The invention provides a kind of quantization loop fluidized-bed combustion boiler coal supply thermal release time system and method, described system comprises modeling method of least squares support module, data decimation and pretreatment module, coal supply thermal release time optimizing module and DCS system and database.The historical data of unit operation is transferred to described data decimation and pretreatment module by described DCS system and database, and described data decimation and pretreatment module are according to the training data of historical data Selection Model and predicted data; The model training data choosing out and predicted data are transferred to described modeling method of least squares support module by described data decimation and pretreatment module, and described modeling method of least squares support module is used for setting up LSSVM model; The training data of the model received is transferred to described coal supply thermal release time optimizing module by described modeling method of least squares support module, described coal supply thermal release time optimizing module processes the training data received and predicted data, determine optimal dynamic rank group, and then determine the coal supply thermal release time and be transferred to described DCS system and database, to instruct unit operation, control.

Described modeling method of least squares support module is used for setting up minimum algorithm of support vector machine coal supply thermal release time model, and the output of model is dynamic bed temperature value, determines to input the thermal release time in coal-supplying amount according to the precision of model prediction.

The historical data of unit operation is transferred to described data decimation and pretreatment module by described DCS system and database, described data decimation and pretreatment module are according to the training data of historical data Selection Model and predicted data, and the training data of the model choosing out is transferred to described modeling method of least squares support module by described data decimation and pretreatment module.

Described modeling method of least squares support module is used for setting up LSSVM model, and the training data of the model received is transferred to described coal supply thermal release time optimizing module by described modeling method of least squares support module.

Described coal supply thermal release time optimizing module processes the training data received and predicted data, determines optimal dynamic rank group, and then determines the coal supply thermal release time and be transferred to described DCS system and database.

The present invention also provides a kind of method of quantization loop fluidized-bed combustion boiler coal supply thermal release time, comprises the following steps:

Step S1, utilizes described modeling method of least squares support module construction least square method supporting vector machine model;

Step S2, utilizes described data decimation and pretreatment module, chooses least square method supporting vector machine Algorithm for Training data in modeling method of least squares support module;

Step S3, the least square method supporting vector machine model that described coal supply thermal release time optimizing module is set up according to step S1, the coal supply thermal release time optimal value in the training data determine described step S2 is chosen.

Utilize the model of described least square method supporting vector machine (leastsquaressupportvectormachine, LSSVM) MBM structure expressed by (1).

y(k)＝f[y(k-1),......,y(k-m)；x ₀(k),......,x ₀(k-n ₀)；

x ₁(k),......,x ₁(k-n ₁)；x ₂(k),......,x ₂(k-n ₂)](1)

Wherein said x (k) is the input quantity of model, described x ₀(k) ..., x ₀(k-n ₀), x ₁(k) ..., x ₁(k-n ₁) and described x ₂(k) ..., x ₂(k-n ₂) be respectively corresponding sampling instant in Circulating Fluidized Bed Boiler and determine the coal-supplying amount of bed temperature, primary air flow and secondary air flow; Described y (k) exports for current bed temperature; Described y (k-1) ..., y (k-m) represents that history bed temperature exports, described m, n ₀, n ₁, n ₂represent the dynamic order of history bed temperature, the dynamic order of coal-supplying amount, the dynamic order of primary air flow and the dynamic order of secondary air flow respectively, characterize the dynamic of bed temperature, span is relevant with the sampling time ts in data decimation and pretreatment module, and the precision determination coal supply thermal release time according to model prediction is n ₀× ts.

LSSVM theory of algorithm principle:

Data set { the X of given N number of model sample _i, y _i} _{i=1 ... N}, wherein i-th sample is input as X _i∈ R ^k(k is input vector dimension), i-th sample exports y _i∈ R.

First, with a Nonlinear Mapping Φ () by the input space R of sample ^kbe mapped to feature space then, in this high-dimensional feature space, optimal decision function is constructed finally, take structural risk minimization as principle Confirming model parameter ω, b.

Utilize structural risk minimization, choose the quadratic term that loss function is error, optimization problem can be described as the problem solved below.

In formula: ω is weight vector, Φ () is mapping function, and ξ i is the predicated error of model to training sample, and c is penalty coefficient.Lagrange method solving-optimizing problem is utilized to obtain

wherein a=[a ₁, a ₂..., a _n] be Lagrange multiplier.According to optimal conditions can obtain:

$\underset{i=1}{\overset{N}{\Σ}}{a}_i=0---\left(i3\right)$

2cξ _i＝a _i(i4)

Formula (i2) (i3) is substituted into formula (i5):

$y_i=\underset{j=1}{\overset{N}{\Σ}}\left(a_{j}K\right(x_i,x_i\left)\right)+b+\frac{1}{2c}{a}_i---\left(i6\right)$

By formula (i3) and (i6) synthesizing linear equation as follows:

$\left[\begin{array}{cc}0& {\stackrel{\→}{1}}^T\\ \stackrel{\→}{1}& \mathrm{\Ω}+V_C\end{array}\right]\left[\begin{array}{c}b\\ a\end{array}\right]=\left[\begin{array}{c}0\\ y\end{array}\right]$

Wherein y=[y ₁..., y _n] ^t, 1=[1 ..., 1] ^t, α=[a ₁..., a _n] ^t, V _c=diag{1/c}, Ω={ Ω _ij| i, j=1,2 ..., N},

the LSSVM model finally obtaining Function Estimation is

$f\left(x\right)=\underset{1}{\overset{N}{\Σ}}{\mathrm{\α}}_{i}K(x,x_i)+b---\left(i7\right)$

In the process determining (1), described LSSVM algorithm adopts Gaussian radial basis function (radialbasisfunction, RBF) core, namely

K(x,x _i)＝exp(-||x-x _i|| ²/σ ²)

LSSVM algorithm and RBF kernel function comprise two unknown parameter c and σ, utilize grid data service and cross validation to carry out.First set the Candidate Set of c and σ be more open grid (c1, σ 1) ..., (cl, σ l) }, carry out cross validation with the node in grid, obtain the grid node corresponding to least error.Then in certain scope class, construct thinner grid, again with the node in grid for parameter is tested, finally obtain the value of optimum c and σ.The initial value of c gets 20, and scope is that the initial value of 0 ~ 200, σ gets 0.3, and scope is 0 ~ 5.

Determine sampling time ts, unit: second, according to the sampling time, described data decimation and pretreatment module obtain the historical data of unit operation by DCS system and database, comprise time point, unit load, coal-supplying amount, primary air flow, secondary air flow, bed temperature value.Due to Circulating Fluidized Bed Boiler delay, inertia and accumulation of heat larger, and these characteristics have certain difference under different load section, and in order to ensure forecasting accuracy, a point load section sets up LSSVM model, generally can by 20 ~ 100WM subregion.History data abundant under selecting different load section, the dynamic rank of each input quantity are chosen and are determined, as the training data of LSSVM algorithm by described coal supply thermal release time optimizing module.

Described coal supply thermal release time optimizing module is carried out optimal dynamic rank group and is extracted, and determines the coal supply thermal release time of corresponding load section, is transferred to described DCS system and database.About determining m, n ₀, n ₁, n ₂the process of Search Range, in conjunction with Circulating Fluidized Bed Boiler characteristic and ts size, the scope of m is 0 ~ 360/ts, described n ₀scope be 0 ~ 900/ts, described n ₁scope be 0 ~ 240/ts, described n ₂scope be 0 ~ 300/ts, be spaced apart 1.

The objective function optimized is the root-mean-square error RMSE (rootmeansquarederrors) of predicted data, RMSE is used to weigh observed reading with the deviation between true value, it is the square root of the quadratic sum observation frequency n ratio of observed reading and true value deviation, namely

$RMSE=\sqrt{\underset{i=1}{\overset{n}{\Σ}}{({\hat{y}}_i-y_i)}^2/n}$

The method of traversal is all adopted to carry out under each load section of optimizing part, namely the root-mean-square error of dynamic rank combination is often organized under calculating corresponding load section, choose the minimum combination of root-mean-square error as optimizing result, if root-mean-square error is less than 1 described in optimizing result, by corresponding for optimizing result m, n ₀, n ₁, n ₂be worth the dynamic rank group optimal value as corresponding load section, will n be worth ₀× ts as the coal supply thermal release time of corresponding load section, otherwise chooses more data from described data decimation and pretreatment module, continues to optimize.

The present invention passes through actual operating data, the delaying of Circulating Fluidized Bed Boilers, inertia and accumulation of heat are taken into full account, under different load section, under different operating condition, quantize the coal supply thermal release time, reach the beneficial effect that the safe operation, quick load change and the control system that are conducive to unit is optimized.

In order to explain the present invention further, be described further below in conjunction with specific embodiment.

embodiment 1

Using the subcritical resuperheat unit boiler of Datang CFB300MW as research object, boiler size is 1100t/h, without external bed heat.Data decimation and pretreatment module acquire continuous ten days 14400 groups of service datas (called after data group 1), get one day 1440 groups of service data (called after data group 2), sampling time is 1 minute, the scope tentatively choosing history bed temperature data order m is [0,6], coal-supplying amount order n ₀scope be [0,15], primary air flow order n ₁scope be [0,4], secondary air flow order n ₂scope be [0,5].

3 sections will be divided in data group 1 as training data according to load: load section 1 is 150MW ~ 200MW in coal supply thermal release time optimizing module; Load section 2 is 200MW ~ 250MW; Load section 3 is 250MW ~ 300MW, and in data group 2, the data of corresponding load section are as forecast test data.

By the data training pattern of load section corresponding in data group 1, with the root-mean-square error RMSE of predicted data for the objective function optimized carries out optimizing, the method of traversal is all adopted to carry out under each load section, namely the root-mean-square error of dynamic rank combination is often organized under calculating corresponding load section, choose the minimum combination of root-mean-square error as optimizing result, if root-mean-square error is less than 1 described in optimizing result, by corresponding for optimizing result m, n ₀, n ₁, n ₂be worth the dynamic rank group optimal value as corresponding load section, will n be worth ₀× ts is as the coal supply thermal release time of corresponding load section, otherwise from described data decimation and pretreatment module, choose more data, continuation is optimized.

embodiment 2

Using the overcritical CFB unit boiler of certain 600MW of Dongfang Boiler Factory manufacture as research object, band external bed heat.Data decimation and pretreatment module acquire continuous eight days 17280 groups of service datas (called after data group 1), get one day 2160 groups of service data (called after data group 2), sampling time is 40 seconds, the scope tentatively choosing history bed temperature data order m is [0,9], coal-supplying amount order n ₀scope be [0,22], primary air flow order n ₁scope be [0,6], secondary air flow order n ₂scope be [0,7].

3 sections will be divided in data group 1 as training data according to load: load section 1 is 300MW ~ 400MW in coal supply thermal release time optimizing module; Load section 2 is 400MW ~ 500MW; Load section 3 is 500MW ~ 600MW, and in data group 2, the data of corresponding load section are as predicted data.

By the data training pattern of load section corresponding in data group 1, with the root-mean-square error RMSE of predicted data for the objective function optimized carries out optimizing, the method of traversal is all adopted to carry out under each load section, namely the root-mean-square error of dynamic rank combination is often organized under calculating corresponding load section, choose the minimum combination of root-mean-square error as optimizing result, if root-mean-square error is less than 0.8 described in optimizing result, by corresponding for optimizing result m, n ₀, n ₁, n ₂be worth the dynamic rank group optimal value as corresponding load section, will n be worth ₀× ts is as the coal supply thermal release time of corresponding load section, otherwise from described data decimation and pretreatment module, choose more data, continuation is optimized.

Claims (10)

1. the system of quantization loop fluidized-bed combustion boiler coal supply thermal release time, is characterized in that, described system comprises:

Modeling method of least squares support module;

Data decimation and pretreatment module;

Coal supply thermal release time optimizing module; And

DCS system and database,

Wherein, described DCS system is connected with described pretreatment module with described data decimation with database, described data decimation and pretreatment module and described modeling method of least squares support module are bi-directionally connected, described modeling method of least squares support module and described coal supply thermal release time optimizing model calling, described coal supply thermal release time optimizing module processes the training data selected and predicted data, determine the coal supply thermal release time of corresponding load section, be transferred to described DCS system and database.

2. system according to claim 1, it is characterized in that, described modeling method of least squares support module is used for setting up minimum algorithm of support vector machine coal supply thermal release time model, the output of model is dynamic bed temperature value, determines to input the thermal release time in coal-supplying amount according to the precision of model prediction.

3. system according to claim 1, it is characterized in that, the historical data of unit operation is transferred to described data decimation and pretreatment module by described DCS system and database, described data decimation and pretreatment module are according to the training data of historical data Selection Model and predicted data, and the model training data choosing out and predicted data are transferred to described modeling method of least squares support module by described data decimation and pretreatment module.

4. system according to claim 1, it is characterized in that, described modeling method of least squares support module is used for setting up LSSVM model, and the model training data received and predicted data are transferred to described coal supply thermal release time optimizing module by described modeling method of least squares support module.

5. system according to claim 1, it is characterized in that, described coal supply thermal release time optimizing module processes the model training data received and predicted data, determines optimal dynamic rank group, and then determines the coal supply thermal release time and be transferred to described DCS system and database.

6. the method for system quantifies Coal Feeding in Circulating Fluidized Bed Furnace thermal release time according to claim 1, is characterized in that, said method comprising the steps of:

Step S1, utilizes described modeling method of least squares support module construction least square method supporting vector machine model;

Step S2, utilizes described data decimation and pretreatment module, chooses least square method supporting vector machine Algorithm for Training data in modeling method of least squares support module;

Step S3, the least square method supporting vector machine model that described coal supply thermal release time optimizing module is set up according to step S1, the coal supply thermal release time optimal value in the training data determine described step S2 is chosen.

7. method according to claim 6, is characterized in that, described step S1 least square method supporting vector machine model is:

y(k)＝f[y(k-1),......,y(k-m)；x ₀(k),......,x ₀(k-n ₀)；

x ₁(k),......,x ₁(k-n ₁)；x ₂(k),......,x ₂(k-n ₂)]

Build in described least square method supporting vector machine model, adopt Gaussian radial basis function core, namely

K(x,x _i)＝exp(-||x-x _i|| ²/σ ²)

Wherein, described x (k) is the input quantity of model, described x ₀(k) ..., x ₀(k-n ₀), x ₁(k) ..., x ₁(k-n ₁) and described x ₂(k) ..., x ₂(k-n ₂) be respectively corresponding sampling instant in Circulating Fluidized Bed Boiler and determine the coal-supplying amount of bed temperature, primary air flow and secondary air flow; Described y (k) exports for current bed temperature; Described y (k-1) ..., y (k-m) represents that history bed temperature exports, described m, n ₀, n ₁, n ₂represent the dynamic order of history bed temperature, the dynamic order of coal-supplying amount, the dynamic order of primary air flow and the dynamic order of secondary air flow respectively, the precision determination coal supply thermal release time according to model prediction is n ₀be multiplied by data sampling time ts; Described σ is a location parameter.

8. method according to claim 6, it is characterized in that, build described least square method supporting vector machine model, adopt least square support vector algorithm, described least square method supporting vector machine algorithm and described Gaussian radial basis function core comprise two unknown parameter c and σ, utilize grid data service and cross validation to carry out.

9. method according to claim 6, is characterized in that, described step S1 also comprises following sub-step:

The Candidate Set of sub-step S1.1, setting c and σ be more open grid (c1, σ 1) ..., (cl, σ l) }, carry out cross validation with the node in grid, obtain the grid node corresponding to least error;

Sub-step S1.2, constructs new grid according to the grid node that above-mentioned sub-step S1.1 obtains, and with the node in grid for parameter is tested, obtains the value of optimum c and σ.

10. method according to claim 6, is characterized in that, described step S2 comprises following sub-step:

Sub-step S2.1, determines sampling time ts, unit: second, obtains historical data from described DCS system and database, comprises time point, unit load, coal-supplying amount, primary air flow, bed temperature value;

Sub-step S2.2, according to the historical data that sub-step S2.1 obtains, a point load section sets up described least square method supporting vector machine model, and subregion scope is between 20 ~ 100WM;

Sub-step S2.3, according between above-mentioned sub-step S2.2 region of differentiation, under selecting different load section, a certain amount of history data is as training data.