CN105930929A - Coal-fired power plant coal low calorific value soft measurement method based on PCA-SVM - Google Patents

Abstract

The present invention discloses a method for predicting a coal-fired power plant coal low calorific value. The method relies on a field measuring instrument, a thermal power unit DCS, a plant level monitoring information system (SIS), and a computer system which carries out software calculation. The software uses the DCS real-time process data stored in the SIS to carry out soft measurement of a coal low calorific value. The method comprises the specific steps of determining auxiliary variables associated with a coal calorific value, using a principal component analysis (PCA) method to integrate multiple auxiliary variables as multiple unrelated comprehensive variables, taking the multiple principal comprehensive variables as the input variables of a low calorific value support vector machine prediction model, and using a genetic algorithm (GA) to find a suitable penalty parameter c and a kernel function parameter g to build a best prediction model.

Description

The coal-fired power plant Lower heat value flexible measurement method of Based PC A-SVM

Technical field

The present invention relates to a kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value.

Background technology

In China's fired power generating unit, coal is the primary raw material of boiler combustion, and coal cost account for the 80% of totle drilling cost, research Ature of coal factor is most important on the impact of boiler combustion.Boiler design include combustion system main apparatus structure parameter, type selecting, The designs of technical specification etc. all be unable to do without coal fuel heating value parameter.Rich coal resources in China, but various places coal varitation is relatively big, And colm is in the majority, and the as-fired coal calorific value of station boiler is generally declining in recent years, thus causes a lot of station boiler Burning coal universal off-design coal, corresponding operating condition off-design operating mode, have a strong impact on the warp of unit operation Ji property and safety, therefore it is necessary that in real time coal-fired calorific value being carried out detection monitors, and according to the practical situation of as-fired coal calorific value Boiler operatiopn is made corresponding adjustment.At present, power plant the most all uses the mode of measurement method to measure fire coal by oxygen bomb instrument Calorific value, the chemical examination time is long, and analysis result discreteness over time and space is big, it is difficult to real-time instruction boiler operatiopn is excellent Change and adjust.Empirical formula method can quickly estimate coal-fired calorific value, but precision is not ideal enough, and scene needs to install simultaneously A lot of measuring points, make troubles to engineer applied.Support vector machine, as a kind of forecasting tool, is built upon statistical learning reason The VC dimension of opinion is theoretical with on Structural risk minization basis, and it is solving small sample, non-linear and high dimensional pattern knowledge Many distinctive advantages are shown in not.On the one hand soft-measuring technique based on support vector machine can make full use of power plant Some data platforms, under relatively low technical costs, automatically search for study from substantial amounts of data by data mining and hide Special relationship in data, thus real-time and the accuracy of measurement of caloric value can be improved to a certain extent.The present invention selects Taken the auxiliary variable that multiple and coal-fired calorific value is relevant, utilize simultaneously genetic algorithm combine cross validation thought optimization support to Amount machine modeling parameters, sets up the on-line prediction model of coal-fired Lower heat value based on support vector machine.

Summary of the invention

Goal of the invention: the goal of the invention of the present invention is to disclose one to fire for fired power generating unit for the deficiencies in the prior art The Forecasting Methodology of coal Lower heat value, by obtaining nearest a period of time auxiliary variable in the historical data base of power plant SIS Data, through the preliminary pretreatment such as Error processing normalization, find out the pivot that can reflect initial data through pivot analysis Analyze model, then penalty parameter c and the kernel functional parameter g of genetic algorithm optimization are set up thermal motor as modeling parameters The support vector machine soft-sensing model of the coal-fired Lower heat value of group.

Technical scheme: for solving above-mentioned technical problem, it is a kind of for fired power generating unit fire coal Lower heat value that the present invention provides Forecasting Methodology, comprises the following steps:

(1) auxiliary variable relevant to coal-fired calorific value and leading variable are determined: auxiliary variable has unit load, total coal Amount, total blast volume, primary air flow, primary air pressure, feedwater flow, oxygen amount, main steam recept the caloric, reheated steam recepts the caloric, Exhaust gas temperature, the most coal-fired Lower heat value of leading variable；

(2) historical data base from SIS obtains nearest 1 hour interior boiler load, total coal amount, total blast volume, Primary air flow, primary air pressure, feedwater flow, oxygen amount, exhaust gas temperature data, and data are carried out Error processing, obtain Data acquisition system is:

D=(d₁,d₂,…,d_p)；

Wherein: p is the total number of above-mentioned auxiliary variable；

(3) historical data base from SIS obtains nearest 1 hour interior main steam pressure and temperature, reheated steam Inlet and outlet pressure and temperature and feed pressure and temperature data, and data are carried out Error processing, then calculate phase Main steam enthalpy, the reheated steam answered import and export enthalpy and boiler feedwater enthalpy, show that corresponding steam caloric receptivity and reheated steam are inhaled Heat；

Steam caloric receptivity, reheated steam caloric receptivity computing formula are as follows:

q₁=h_gq-h_gs；

q₂=h_zr2-h_zr1；

Wherein: q₁Recept the caloric for main steam, kJ/kg；h_gqFor main steam enthalpy, kJ/kg；h_gsFor Enthalpy of Feed Water； q₂Recept the caloric for reheated steam, kJ/kg；h_zr1、h_zr2It is respectively reheated steam and imports and exports enthalpy, kJ/kg.

(4) combination (2) and (3) show that the data acquisition system of all auxiliary variables is:

X=(D, q₁,q₂)=(d₁,d₂,…,d_p,q₁,q₂)；

In order to avoid the impact of variable difference dimension, above-mentioned data acquisition system is normalized, after obtaining standardization Data acquisition system:

X '=(D ', q '₁,q’₂)=(d '₁,d’₂,…,d’_p,q’₁,q’₂)。

(5) found out by pca method (PCA) the Principal Component Analysis Model Y of initial data X can be characterized, and by Y As training set；

(6) use genetic algorithm that training set data is trained, find the penalty parameter c that supporting vector machine model is optimum With kernel functional parameter g；

(7) use step (6) optimized parameter that obtains and data set Y and Lower heat value data, use and support vector Machine method sets up the soft-sensing model of coal-fired Lower heat value；

(8) data are carried out Error processing, utilize above-mentioned foundation by the online each auxiliary variable data of acquisition in real time PCA-SVM model, obtains coal-fired Lower heat value.

Further, the acquisition time in the data acquisition system in step (2) (3) is spaced apart 30s.

Further, the Error processing described in step (2) (3) includes the process of gross error and random error.

The process of gross error is according to Pauta criterion, and its mathematical method is expressed as follows: set sample collection data as y₁,y₂,…,y_n, then calculated standard deviation by Bessel formula, such as following formula.

$\mathrm{\σ}={(\underset{i=1}{\overset{n}{\Σ}}{v}_i^2/n-1)}^{1/2}={\{\[\underset{i=1}{\overset{n}{\Σ}}{y}_i^2-\left(\underset{i=1}{\overset{n}{\Σ}}{y}_i^2\right)/n\]/(n-1)\}}^{1/2}$

Wherein: n is for gathering data amount check；For gathering the arithmetic average of data；v_iFor gathering data and mean deviation, i.e.

If a certain sample data y_kDeviation v_k(1≤k≤n) meets | v_k| ＞ 3 σ, the most now data are unreasonable, it should pick Remove；

The process of random error uses five point Linear smoothing techniques:

If the volume sample collection data of a certain auxiliary variable are { x₁,x₂,…,x_n, x_i=s_i+n_i, s in formula_iFor true value, n_iFor Noise, then the data after five point Linear smoothing processing are

$y_i=\underset{r=-q}{\overset{q}{\Σ}}{a}_r{x}_{i+r}=\underset{r=-q}{\overset{q}{\Σ}}{a}_r(s_{i+r}+n_{i+r})=\underset{r=-q}{\overset{q}{\Σ}}{a}_r{s}_{i+r}+\underset{r=-q}{\overset{q}{\Σ}}{a}_r{n}_{i+r}=\stackrel{\‾}{s_i}+\stackrel{\‾}{n_i},$

In formula: r is the integer between-q to q, i.e. r=-q ..., 0 ..., q；{ ar} is one group of weighted value, meets

Further, asking for enthalpy by temperature and pressure in step (3) is to utilize WASPCN (water and vapor quality Software for calculation, freeware, can download on the net) try to achieve.

Further, the normalized in step (4) is interval by initial data scale conversion to [0,1], and method is as follows: If the collection data of a certain auxiliary variable are X={x₁,x₂,…,x_i,…,x_n, then data X after standardization ' be:

$X^{,}=\frac{X-min\left(X\right)}{\max \left(X\right)-min\left(X\right)}$

Wherein: min (X) is the minima gathering data X；Max (X) is the maximum gathering data X.

Further, step (7) is utilize LIBSVM workbox to set up supporting vector machine model.

Further, in step (7), supporting vector machine model uses Radial basis kernel function:

Beneficial effect: the present invention possesses advantages below in terms of existing technologies:

(1) present invention is without complicated hardware device, cheap, and traditional coal analysis instrument price is high.

(2) present invention is the flexible measurement method of a kind of calorific value, analyzes method speed more compared to traditional laboratory chemical Hurry up, be a kind of on-line analysis.

(3) present invention measures calorific value precision height, meets actual demands of engineering, and traditional measurement of caloric value error is bigger.

Accompanying drawing explanation

Fig. 1 is data study and the training flow chart of fire coal Lower heat value Forecasting Methodology of the present invention.

Fig. 2 is that the present invention utilizes genetic algorithm to draw the algorithm flow chart of optimal parameter c and g.

Fig. 3 is the fitness curve in embodiment in searching process.

Fig. 4 is modeling and the prediction effect of calorific value model in embodiment.

Detailed description of the invention

Below in conjunction with the accompanying drawings the present invention is further described.

As it is shown in figure 1, on-the-spot DCS sampled data is stored in the historical data base of plant level supervisory information system (SIS), obtain The data of auxiliary variable relevant to coal-fired calorific value in nearest 1 hour, auxiliary variable mainly include boiler load, total coal amount, Total blast volume, primary air flow, primary air pressure, feedwater flow, oxygen amount, exhaust gas temperature data, two other auxiliary variable master Steam caloric receptivity and reheated steam caloric receptivity by main steam pressure and temperature, reheated steam inlet and outlet pressure and temperature and Feed pressure and temperature data calculate.Then data being carried out pretreatment, the process of gross error is accurate according to La Yida Then, deviation is rejected | v_k| sample data y of ＞ 3 σ_k.The process of random error uses five point Linear smoothing techniques.Data are returned The Principal Component Analysis Model Y that can characterize initial data X is found out in one change by pca method (PCA) after processing, and will Y, as training set, uses LIBSVM workbox to set up supporting vector machine model, and model uses RBFMeanwhile, genetic algorithm is used to find the penalty parameter c that supporting vector machine model is optimum With kernel functional parameter g, set up optimal Lower heat value soft-sensing model, improve model prediction accuracy.

As in figure 2 it is shown, in order to ensure that SVM prediction model has higher accuracy rate, genetic algorithm can be passed through Find the globally optimal solution under CV meaning, and if have c and g of many groups corresponding to the highest predictablity rate, Choose can that group c and g that wherein parameter c is minimum as optimal parameter, do so is mainly for avoiding too high c The generation of learning state can be caused.The fitness function of this algorithm is the K-CV method corresponding for c and g with a certain group Predictablity rate.

Example:

Practicality herein in conjunction with Anhui power plant test data analyzer the inventive method.Table 1 is to process through data and inhale Calorific value modeling data after heat Calculation.

Table 1 calorific value modeling data

Make 1～2,4～6,8～10 group as training data, 3,7,11 groups as test data.Utilize the inventive method, Obtaining the optimal parameter of Lower heat value model to be respectively as follows: c and take 12, g takes 0.215, and Fig. 3 is that the adaptation of parameter optimization is write music Line chart, because what fitness function selected is the opposite number of mean square error, therefore the biggest error of fitness is the least, can from figure To find out that final mean square error is stable 1.063.

Now set up optimal supporting vector machine model by the parameter after optimizing, and by this model to 3,7,11 groups Coal-fired Lower heat value is predicted, and result is as shown in Figure 4.As eight groups of data of training data, its actual value and model Predictive value closely, illustrates that training error is relatively small.In figure, three groups of data of labelling circle are as the test of this model Data, from figure it can be seen that actual value and predictive value relatively, maximum error is the 11st group, but this group is definitely missed Difference is 178MJ/kg, and relative error is less than 1%, therefore the inventive method of this Lower heat value prediction disclosure satisfy that the reality of engineering Border requirement.

The above is only the preferred embodiment of the present invention, it should be pointed out that: for those skilled in the art For, under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications are also Should be regarded as protection scope of the present invention.

Claims (7)

1. the Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that comprise the following steps:

(1) auxiliary variable relevant to coal-fired calorific value and leading variable are determined: auxiliary variable has unit load, total coal Amount, total blast volume, primary air flow, primary air pressure, feedwater flow, oxygen amount, main steam recept the caloric, reheated steam recepts the caloric, Exhaust gas temperature, the most coal-fired Lower heat value of leading variable；

(2) historical data base from SIS obtains nearest 1 hour interior boiler load, total coal amount, total blast volume, Primary air flow, primary air pressure, feedwater flow, oxygen amount, exhaust gas temperature data, and data are carried out Error processing, obtain Data acquisition system is:

D=(d₁,d₂,…,d_p)；

Wherein: p is the total number of above-mentioned auxiliary variable；

(3) historical data base from SIS obtains nearest 1 hour interior main steam pressure and temperature, reheated steam Inlet and outlet pressure and temperature and feed pressure and temperature data, and data are carried out Error processing, then calculate phase Main steam enthalpy, the reheated steam answered import and export enthalpy and boiler feedwater enthalpy, show that corresponding steam caloric receptivity and reheated steam are inhaled Heat；

Steam caloric receptivity, reheated steam caloric receptivity computing formula are as follows:

q₁=h_gq-h_gs；

q₂=h_zr2-h_zr1；

Wherein: q₁Recept the caloric for main steam, kJ/kg；h_gqFor main steam enthalpy, kJ/kg；h_gsFor Enthalpy of Feed Water； q₂Recept the caloric for reheated steam, kJ/kg；h_zr1、h_zr2It is respectively reheated steam and imports and exports enthalpy, kJ/kg.

(4) combination (2) and (3) show that the data acquisition system of all auxiliary variables is:

X=(D, q₁,q₂)=(d₁,d₂,…,d_p,q₁,q₂)；

In order to avoid the impact of variable difference dimension, above-mentioned data acquisition system is normalized, after obtaining standardization Data acquisition system:

X '=(D ', q '₁,q’₂)=(d '₁,d’₂,…,d’_p,q’₁,q’₂)。

(5) found out by pca method (PCA) the Principal Component Analysis Model Y of initial data X can be characterized, and by Y As training set；

(6) use genetic algorithm that training set data is trained, find the penalty parameter c that supporting vector machine model is optimum With kernel functional parameter g；

(7) use step (6) optimized parameter that obtains and data set Y and Lower heat value data, use and support vector Machine method sets up the soft-sensing model of coal-fired Lower heat value；

(8) data are carried out Error processing, utilize above-mentioned foundation by the online each auxiliary variable data of acquisition in real time PCA-SVM model, obtains coal-fired Lower heat value.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that The acquisition time in data acquisition system in step (2) (3) is spaced apart 30s.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that Error processing described in step (2) (3) includes the process of gross error and random error.

The process of gross error is according to Pauta criterion, and its mathematical method is expressed as follows: set sample collection data as y₁,y₂,…,y_n, then calculated standard deviation by Bessel formula, such as following formula.

$\mathrm{\σ}={(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{v}_i^2/n-1)}^{1/2}={\{\[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i^2-\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i^2\right)/n\]/(n-1)\}}^{1/2}$

Wherein: n is for gathering data amount check；For gathering the arithmetic average of data；v_iFor gathering data and mean deviation, i.e.(i=1,2 ..., n).

If a certain sample data y_kDeviation v_k(1≤k≤n) meets | v_k| ＞ 3 σ, the most now data are unreasonable, it should pick Remove；

The process of random error uses five point Linear smoothing techniques:

If the volume sample collection data of a certain auxiliary variable are { x₁,x₂,…,x_n, x_i=s_i+n_i, s in formula_iFor true value, n_iFor Noise, then the data after five point Linear smoothing processing are

$y_i=\underset{r=-q}{\overset{q}{\mathrm{\Σ}}}{a}_r{x}_{i+r}=\underset{r=-q}{\overset{q}{\mathrm{\Σ}}}{a}_r(s_{i+r}+n_{i+r})=\underset{r=-q}{\overset{q}{\mathrm{\Σ}}}{a}_r{s}_{i+r}+\underset{r=-q}{\overset{q}{\mathrm{\Σ}}}{a}_r{n}_{i+r}=\stackrel{\‾}{s_i}+\stackrel{\‾}{n_i},$

In formula: r is the integer between-q to q, i.e. r=-q ..., 0 ..., q；{a_rIt is one group of weighted value, meetQ=5.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that Asking for enthalpy by temperature and pressure in step (3) is to utilize WASPCN to try to achieve.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that Normalized in step (4) is interval by initial data scale conversion to [0,1], and method is as follows: sets a certain auxiliary and becomes The collection data of amount are X={x₁,x₂,…,x_i,…,x_n, then data X after standardization ' be:

$X^{,}=\frac{X-\min \left(X\right)}{\max \left(X\right)-min\left(X\right)}$

Wherein: min (X) is the minima gathering data X；Max (X) is the maximum gathering data X.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that Step (7) is utilize LIBSVM workbox to set up supporting vector machine model.

A kind of Forecasting Methodology for fired power generating unit fire coal Lower heat value, it is characterised in that In step (7), supporting vector machine model uses Radial basis kernel function:

$K(x,x_i)=\exp \{-\frac{\left|\right|x-x_i|{|}^2}{{\mathrm{\σ}}^2}\}.$