CN104832210B - The gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM - Google Patents

Abstract

The invention discloses the gas absolute discharge Forecasting Methodology of a kind of Based PC A-FIG-SVM, belong to the Forecasting Methodology of stope working surface of coal mines gas absolute discharge, comprise the following steps: the collection of (1) gas absolute discharge monitored data and influence factor; (2) principal component modeling is carried out to influence factor data, reconstruct principal component; (3) Fuzzy Information Granulation is carried out to the time series that gas absolute discharge monitored data is formed; (4) the Support vector regression model of granulation data is set up; (5) gas absolute discharge prediction.The present invention proposes a kind of forecast model of new gas absolute discharge variation tendency, the spatial dimension of primary study variation tendency and change, principal component analysis is utilized to extract principal component, effectively reduce the impact of lengthy and jumbled information, reduce the input dimension of SVR model, its design principle is reliable, and Forecasting Methodology is simple, precision of prediction is high, prediction environmental friendliness.

Description

The gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM

Technical field

The present invention relates to the Forecasting Methodology of driving face gas emission, particularly a kind of gas absolute discharge Forecasting Methodology based on principal component analysis (PCA)-Fuzzy Information Granulation (FIG)-SVMs (SVM).

Background technology

In process of coal mining, gas problem is one of major hidden danger of restriction Safety of Coal Mine Production always, and accurately measuring with the absolute outburst amount of real-time monitoring coal mine gas is the important measures ensureing Safety of Coal Mine Production.

At present, domestic and international many scholars have carried out deeply careful research to the forecasting problem of gas, propose a lot of effective Forecasting Methodology, are broadly divided into two classes: a class is linear model, as point source predicted method, principle component regression method, a statistic law etc.; Another kind of is forecast model based on nonlinear combination, as artificial neural network method, chaos forecast method, SVMs method, fuzzy mathematics and very fast learning machine method etc.; These methods respectively have advantage and serve certain facilitation to Forecast of Gas Emission.

Along with country is to the needs of improving constantly of requiring of Safety of Coal Mine Production and enterprise's self-growth, each big-and-middle-sized colliery of China is equipped with Colliery Safety Supervise System all successively, greatly improves mine safety production level and production safety management efficiency.But these monitor datas are only a kind of records of current operating state, also lack the development trend to monitor data within following a period of time.In the prior art, Wu Zhaofa etc. disclose a kind of gas density trend forecasting method on " industrial and mineral automation " periodical the 40th volume the 12nd phase in 2014, paper is called: based on the gas density trend prediction of the trapezoidal Fuzzy Information Granulation of interpolation, but because Forecasting Methodology is only based on gas density initial data, do not consider mining conditions, the effect of the influence factors such as coal seam conditions, in mining process, the influence factor affecting gas emission is among constantly change, face gas outburst amount is made to there is very large uncertainty, therefore the forecast model proposing a kind of new gas absolute discharge variation tendency is necessary, the spatial dimension of primary study variation tendency and change.

Summary of the invention

The object of the invention is for overcoming above-mentioned the deficiencies in the prior art, the gas absolute discharge Forecasting Methodology of a kind of Based PC A-FIG-SVM is provided, the method can improve Forecast of Gas Emission precision, the spatial dimension of research variation tendency and change, make up original Forecast of Gas Emission defect, for Safety of Coal Mine Production provides foundation.

For achieving the above object, the present invention adopts following technical proposals:

A gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM, comprises the following steps:

(1) collection of gas absolute discharge monitored data and influence factor; Wherein, influence factor comprise coal seam thickness, coal seam gas-bearing capacity, coal seam spacing, day fltting speed and average daily output;

(2) carry out principal component modeling to influence factor data, reconstruct principal component, concrete steps are as follows:

1. influence factor data are normalized, obtain sample set matrix X;

2. sample set matrix X following formula is transformed to correlation matrix, obtains principal component matrix R:

R＝(r _ij) _p×p

And $r_{ij}=\frac{1}{n}\underset{d=1}{\overset{n}{\Σ}}(x_{di}-{\stackrel{\‾}{x}}_i)(x_{di}-{\stackrel{\‾}{x}}_j),(i=1,2,...,p;j=1,2,...,p)$

Wherein: x _dibe the numerical value of i-th influence factor d sample; be the average of i-th all sample values of influence factor; x _djfor the numerical value of a jth influence factor d sample; for the average of all sample values of a jth influence factor; N is number of samples; P is influence factor number; r _ijit is the index of correlation of i-th influence factor and a jth influence factor;

3. obtain characteristic value, principal component contributor rate and contribution rate of accumulative total according to principal component matrix R, determine principal component number m according to contribution rate of accumulative total > 90%, and set up principal component model by following formula:

F _k＝α _1kX ₁+α _2kX ₂+...+α _pkX _p(k＝1,2,...,m)

Wherein, F _kfor kth principal component, the coefficient vector (α in each equation _1k, α _2k..., α _pk) be eigenvalue λ respectively ₁, λ ₂..., λ _mcorresponding unit character vector, X _i(i=1,2 ... p) be the standardized data of i-th influence factor;

(3) Fuzzy Information Granulation is carried out to the time series that gas absolute discharge monitored data is formed; Concrete steps are as follows:

1. partition window: determine the size l being granulated window, by gas absolute discharge sequence Y={y ₁, y ₂..., y _mwith the size l being granulated window for sub-row length is divided into [M/l] height row Y ' _n, n=1,2 ..., M/l, wherein [M/l] represents round numbers forward;

2. on a sub-row window, obscure particle A is set up;

First the form of obscure particle is selected: adopt triangular form obscure particle, its membership function is as follows:

$A(x,a,c,b)=\left\{\begin{array}{cc}0,& x<a\\ \frac{x-a}{c-a},& a\&le;x\&le;c\\ \frac{b-x}{b-c},& c\&le;x\&le;b\\ 0,& x>b\end{array}\right.$

In formula: x is the gas absolute discharge gathered; A and b is support lower limit and the upper limit of obscure particle A respectively; C is the core of obscure particle A;

Then, the parameter a of obscure particle A, b and c is determined: the core c of obscure particle A gets the average of sub-column data collection; Parameter a, b are then by solving the optimization problem described in formula (1):

$\underset{a,b}{MaxQ}=\frac{\underset{i=1}{\overset{l}{\&Sigma;}}A\left(Y_{ni}^{\&prime;}\right)}{measure\left(supp\right(A\left)\right)}=\frac{\underset{i=1}{\overset{l}{\&Sigma;}}A\left(Y_{ni}^{\&prime;}\right)}{b-a}---\left(1\right)$

Wherein, Max represents maximization operation, Y ' _nifor sub-row Y ' _nin element, l be son row Y ' _nlength, A (Y ' _ni) represent Y ' _nmembership function.

3. Fuzzy Information Granulation is carried out to gas absolute discharge;

According to step 2., on [M/l] height row window of Y, set up corresponding obscure particle A respectively, obtain gas absolute discharge and carry out Fuzzy Information Granulation result, obtain 3 Fuzzy Information Granulation Variables Sequences:

$\left\{\begin{array}{c}Low=\&lsqb;l_1^{\&prime;},l_2^{\&prime;},...,l_{\&lsqb;M/l\&rsqb;}^{\&prime;}\&rsqb;\\ R=\&lsqb;r_1,r_2,...,r_{\&lsqb;M/l\&rsqb;}\&rsqb;\\ Up=\&lsqb;u_1,u_2,...,u_{\&lsqb;M/l\&rsqb;}\&rsqb;\end{array}\right.$

Wherein, l ' ₁, l ' ₂..., l _[M/l]for the minimum value of Y after Fuzzy Information Granulation on every height row window; r ₁, r ₂..., r _[M/l]for the average of Y after Fuzzy Information Granulation on every height row window; u ₁, u ₂..., u _[M/l]for the maximum value of Y after Fuzzy Information Granulation on every height row window; Low parameter is the minimum value of Y change, and R parameter is the cardinal principle average level of Y change, and Up parameter is the maximum value of Y change.

(4) set up the Support vector regression model of granulation data, step is as follows:

1. pretreatment is carried out to reconstruct principal component: according to the size l of granulation window, respectively each is reconstructed main one-tenth and be divided into [M/l] height row, the Principal component that the reconstruct principal component average getting each height row arranges as this son.

2. set up sample set: with pretreated reconstruct number of principal components certificate and 3 Fuzzy Information Granulation Variables Sequence data composition sample sets, and the sample choosing 20% is as test samples, all the other are as training sample;

3. the Support vector regression model of Fuzzy Information Granulation Variables Sequence Low, R and Up is set up: first utilize gravitation searching algorithm to carry out optimizing to the penalty parameter c of Support vector regression model and nuclear parameter σ, wherein Selection of kernel function RBF; Then the optimized parameter sought is utilized to train training sample; set up the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up respectively; and with test samples to model testing; if average relative error < 10%; think that model is reliable; can be used to prediction, otherwise, principal component modeling reconstruct principal component again.

(5) gas absolute discharge prediction; concrete grammar is; first the principal component model applying foundation extracts principal component and carries out pretreatment; be entered into the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up that step (4) is set up, the maximum value of gas emission, minimum value and average can be doped.

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

(1) the present invention is using gas emission influence factor as the input parameter setting up FIG-SVM model, and replace using gas emission time series monitored data or simple time as input parameter in traditional F IG-SVM model, the model of foundation is more reliable.

(2) utilize principal component analysis to carry out dimension-reduction treatment to gas emission influence factor, effectively reduce the impact of lengthy and jumbled information, reduce mode input dimension, greatly improve model learning efficiency and precision.

Accompanying drawing explanation

Fig. 1 is flow chart of the present invention.

Detailed description of the invention

Below in conjunction with drawings and Examples, the present invention is further described.

As shown in Figure 1, the gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM, comprises the following steps:

(1) collection of gas absolute discharge monitored data and influence factor; Wherein, influence factor comprise coal seam thickness, coal seam gas-bearing capacity, coal seam spacing, day fltting speed and average daily output.

(2) carry out principal component modeling to influence factor data, reconstruct principal component, concrete steps are as follows

1. influence factor data are normalized, obtain sample set matrix X;

2. sample set matrix X following formula is transformed to correlation matrix, obtains principal component matrix R:

R＝(r _ij) _p×p

And $r_{ij}=\frac{1}{n}\underset{d=1}{\overset{n}{\&Sigma;}}(x_{di}-{\stackrel{\&OverBar;}{x}}_i)(x_{di}-{\stackrel{\&OverBar;}{x}}_j),(i=1,2,...,p;j=1,2,...,p)$

Wherein: x _dibe the numerical value of i-th influence factor d sample; be the average of i-th all sample values of influence factor; x _djfor the numerical value of a jth influence factor d sample; for the average of all sample values of a jth influence factor; N is number of samples; P is influence factor number; r _ijit is the index of correlation of i-th influence factor and a jth influence factor;

3. obtain characteristic value, principal component contributor rate and contribution rate of accumulative total according to principal component matrix R, determine principal component number m according to contribution rate of accumulative total > 90%, and set up principal component model by following formula:

F _k＝α _1kX ₁+α _2kX ₂+...+α _pkX _p(k＝1,2,...,m)

Wherein, F _kfor kth principal component, the coefficient vector (α in each equation _1k, α _2k..., α _pk) be eigenvalue λ respectively ₁, λ ₂..., λ _mcorresponding unit character vector, X _i(i=1,2 ... p) be the standardized data of i-th influence factor;

(3) Fuzzy Information Granulation is carried out to the time series that gas absolute discharge monitored data is formed; Concrete steps are as follows:

1. partition window: determine the size l being granulated window, by gas absolute discharge sequence Y={y ₁, y ₂..., y _mwith the size l being granulated window for sub-row length is divided into [M/l] height row Y ' _n, n=1,2 ..., M/l, wherein [M/l] represents round numbers forward;

2. on a sub-row window, obscure particle A is set up;

First the form of obscure particle is selected: adopt triangular form obscure particle, its membership function is as follows:

$A(x,a,c,b)=\left\{\begin{array}{cc}0,& x<a\\ \frac{x-a}{c-a},& a\&le;x\&le;c\\ \frac{b-x}{b-c},& c\&le;x\&le;b\\ 0,& x>b\end{array}\right.$

In formula: x is the gas absolute discharge gathered; A and b is support lower limit and the upper limit of obscure particle A respectively; C is the core of obscure particle A;

Then, the parameter a of obscure particle A, b and c is determined: the core c of obscure particle A gets the average of sub-column data collection; Parameter a, b are then by solving the optimization problem described in formula (1):

$\underset{a,b}{MaxQ}=\frac{\underset{i=1}{\overset{l}{\&Sigma;}}A\left(Y_{ni}^{\&prime;}\right)}{measure\left(supp\right(A\left)\right)}=\frac{\underset{i=1}{\overset{l}{\&Sigma;}}A\left(Y_{ni}^{\&prime;}\right)}{b-a}---\left(1\right)$

Wherein, Max represents maximization operation, Y ' _nifor sub-row Y ' _nin element, l be son row Y ' _nlength, A (Y ' _ni) represent Y ' _nmembership function.

3. Fuzzy Information Granulation is carried out to gas absolute discharge;

According to step 2., on [M/l] height row window of Y, set up corresponding obscure particle A respectively, obtain gas absolute discharge and carry out Fuzzy Information Granulation result, obtain 3 Fuzzy Information Granulation Variables Sequences:

$\left\{\begin{array}{c}Low=\&lsqb;l_1^{\&prime;},l_2^{\&prime;},...,l_{\&lsqb;M/l\&rsqb;}^{\&prime;}\&rsqb;\\ R=\&lsqb;r_1,r_2,...,r_{\&lsqb;M/l\&rsqb;}\&rsqb;\\ Up=\&lsqb;u_1,u_2,...,u_{\&lsqb;M/l\&rsqb;}\&rsqb;\end{array}\right.$

Wherein, l ' ₁, l ' ₂..., l _[M/l]for the minimum value of Y after Fuzzy Information Granulation on every height row window; r ₁, r ₂..., r _[M/l]for the average of Y after Fuzzy Information Granulation on every height row window; u ₁, u ₂..., u _[M/l]for the maximum value of Y after Fuzzy Information Granulation on every height row window; Low parameter is the minimum value of Y change, and R parameter is the cardinal principle average level of Y change, and Up parameter is the maximum value of Y change.

(4) set up the Support vector regression model of granulation data, step is as follows:

1. pretreatment is carried out to reconstruct principal component: according to the size l of granulation window, respectively each is reconstructed main one-tenth and be divided into [M/l] height row, the Principal component that the reconstruct principal component average getting each height row arranges as this son.

2. set up sample set: with pretreated reconstruct number of principal components certificate and 3 Fuzzy Information Granulation Variables Sequence data composition sample sets, and the sample choosing 20% is as test samples, all the other are as training sample;

3. the Support vector regression model of Fuzzy Information Granulation Variables Sequence Low, R and Up is set up: first utilize gravitation searching algorithm to carry out optimizing to the penalty parameter c of Support vector regression model and nuclear parameter σ, wherein Selection of kernel function RBF; Then the optimized parameter sought is utilized to train training sample; set up the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up respectively; and with test samples to model testing; if average relative error < 10%; think that model is reliable; can be used to prediction, otherwise, principal component modeling reconstruct principal component again.

(5) gas absolute discharge prediction; concrete grammar is; first the principal component model applying foundation extracts principal component and carries out pretreatment; be entered into the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up that step (4) is set up, the maximum value of gas emission, minimum value and average can be doped.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.

Claims (5)

1. the gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM, is characterized in that: comprise the following steps:

(1) collection of gas absolute discharge monitored data and influence factor;

(2) principal component modeling is carried out to influence factor data, reconstruct principal component; Step is as follows:

1. influence factor data are normalized, obtain sample set matrix X;

2. sample set matrix X following formula is transformed to correlation matrix, obtains principal component matrix R:

R＝(r _ij) _p×p

And

Wherein: x _dibe the numerical value of i-th influence factor d sample; be the average of i-th all sample values of influence factor; x _djfor the numerical value of a jth influence factor d sample; for the average of all sample values of a jth influence factor; N is number of samples; P is influence factor number; r _ijit is the index of correlation of i-th influence factor and a jth influence factor;

3. obtain characteristic value, principal component contributor rate and contribution rate of accumulative total according to principal component matrix R, determine principal component number m according to contribution rate of accumulative total > 90%, and set up principal component model by following formula:

F _k＝α _1kX ₁+α _2kX ₂+...+α _pkX _p(k＝1,2,...,m)

Wherein, F _kfor kth principal component, the coefficient vector (α in each equation _1k, α _2k..., α _pk) be eigenvalue λ respectively ₁, λ ₂..., λ _mcorresponding unit character vector, X _i(i=1,2 ... p) be the standardized data of i-th influence factor;

(3) Fuzzy Information Granulation is carried out to the time series that gas absolute discharge monitored data is formed;

(4) the Support vector regression model of granulation data is set up;

(5) gas absolute discharge prediction.

2. the gas absolute discharge Forecasting Methodology of a kind of Based PC A-FIG-SVM according to claim 1, is characterized in that: the influence factor of described step (1) comprise coal seam thickness, coal seam gas-bearing capacity, coal seam spacing, day fltting speed and average daily output.

3. the gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM according to claim 1, is characterized in that: the Fuzzy Information Granulation of described step (3), and step is as follows:

1. partition window: determine the size l being granulated window, by gas absolute discharge sequence Y={y ₁, y ₂..., y _mwith the size l being granulated window for sub-row length is divided into [M/l] height row Y ' _n, n=1,2 ..., M/l, wherein [M/l] represents round numbers forward;

2. on a sub-row window, obscure particle A is set up;

First the form of obscure particle is selected: adopt triangular form obscure particle, its membership function is as follows:

In formula: x is the gas absolute discharge gathered; A and b is support lower limit and the upper limit of obscure particle A respectively; C is the core of obscure particle A;

Then, the parameter a of obscure particle A, b and c is determined: the core c of obscure particle A gets the average of sub-column data collection; Parameter a, b are then by solving the optimization problem described in formula (1):

Wherein, Max represents maximization operation, Y ' _nifor sub-row Y ' _nin element, l be son row Y ' _nlength, A (Y ' _ni) represent Y ' _nmembership function;

3. Fuzzy Information Granulation is carried out to gas absolute discharge;

According to step 2., on [M/l] height row window of Y, set up corresponding obscure particle A respectively, obtain gas absolute discharge and carry out Fuzzy Information Granulation result, obtain 3 Fuzzy Information Granulation Variables Sequences:

Wherein, l ' ₁, l ' ₂..., l _[M/l]for the minimum value of Y after Fuzzy Information Granulation on every height row window; r ₁, r ₂..., r _[M/l]for the average of Y after Fuzzy Information Granulation on every height row window; u ₁, u ₂..., u _[M/l]for the maximum value of Y after Fuzzy Information Granulation on every height row window; Low parameter is the minimum value of Y change, and R parameter is the cardinal principle average level of Y change, and Up parameter is the maximum value of Y change.

4. the gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM according to claim 3, is characterized in that: the Support vector regression model of the foundation granulation data of described step (4), and step is as follows:

1. pretreatment is carried out to reconstruct principal component: according to the size l of granulation window, respectively each is reconstructed main one-tenth and be divided into [M/l] height row, the Principal component that the reconstruct principal component average getting each height row arranges as this son;

2. set up sample set: with pretreated reconstruct number of principal components certificate and 3 Fuzzy Information Granulation Variables Sequence Low, R and Up data composition sample sets, and the sample choosing 20% is as test samples, all the other are as training sample;

3. the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up is set up: first utilize gravitation searching algorithm to carry out optimizing to the penalty parameter c of Support vector regression model and nuclear parameter σ, wherein Selection of kernel function RBF; Then the optimized parameter sought is utilized to train training sample; set up the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up respectively; and with test samples to model testing; if average relative error < 10%; think that model is reliable; can be used to prediction, otherwise, principal component modeling reconstruct principal component again.

5. the gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM according to claim 4; it is characterized in that: the concrete grammar of described step (5) gas absolute discharge prediction is; first the principal component model applying foundation extracts principal component and carries out pretreatment; be entered into the Support vector regression model of 3 Fuzzy Information Granulation Variables Sequences Low, R and Up that step (4) is set up, the maximum value of gas emission, minimum value and average can be doped.