CN103630588B - A kind of fast response method of galvanochemistry firedamp sensor - Google Patents

Abstract

The invention provides a kind of fast response method of galvanochemistry firedamp sensor, be specially: be passed through the some gas concentrations in transient by Quick Acquisition gas density, for setting up discrete grey model DGM(1,1), then utilize gray prediction result to train Gaussian process regression model with the actual gas concentration gathered, finally obtain gas concentration in advance based on trained Gaussian process regression model.The present invention can obtain tested gas concentration fast and accurately, reduces the response time of galvanochemistry firedamp sensor dramatically.

Description

A kind of fast response method of galvanochemistry firedamp sensor

Technical field

The present invention relates to firedamp sensor technical field, be specifically related to a kind of fast response method of galvanochemistry firedamp sensor.

Background technology

Firedamp sensor is used for the methane content of monitoring continuously in progress of coal mining in air, i.e. gas density.Existing firedamp sensor mainly contains the type such as optical profile type and electric chemical formula (carrier catalyst element formula).In current China's coal-mine industry, the most extensively and in a large number use electric chemical formula methane transducer.

Electric chemical formula methane transducer adopts carrier catalyst element to be detecting element.During work, the methane in test environment enters sensor probe air chamber and sensitive element generation chemical reaction with diffusion way through the metal sponge playing buffer action, thus produces the electric signal corresponding with methane concentration.This structure of sensor and principle determine it and have considerable transfer delay.

The dynamic response characteristic of firedamp sensor exponentially approaches measured concentration value gradually.When the methane concentration of gas changes, sensor can not indicate the actual value of methane concentration immediately.Need the process of a gradual change.Until transient process terminates the actual value that could obtain concentration.And the signal processing systems such as single-chip microcomputer also need certain signal transacting and signal transmission time, therefore firedamp sensor also exists certain hysteresis delay time, namely there is operating lag phenomenon.Time delay of all kinds of firedamp sensor (as optical profile type, infrared type, carrier catalyst element formula) is not etc.The time delay of the electric chemical formula methane transducer that current China the most generally adopts is greatly about 10 ~ 30 seconds.

Along with the development of observation and control technology, propose more and more higher requirement to the performance of sensor, sensor must have good dynamic perfromance and quick susceptibility.In the coal production field very high to security requirement, on-the-spot gas density need be obtained quickly and accurately by firedamp sensor, require that firedamp sensor dynamic perfromance is high, and time delay be the smaller the better, namely can respond the change of methane concentration fast.In prior art, mostly be by improving sensitive element itself to improve the response speed of firedamp sensor, and the inherent characteristic of sensitive material itself makes response speed room for promotion less.

Based on Such analysis, the present invention proposes a kind of fast response method of firedamp sensor, and to reduce firedamp sensor time delay, this case produces thus.

Summary of the invention

Object of the present invention, be a kind of fast response method providing galvanochemistry firedamp sensor, it can shorten the time delay of firedamp sensor, improves the dynamic response characteristic of firedamp sensor, to obtain tested gas concentration fast and accurately.

In order to reach above-mentioned purpose, solution of the present invention is:

A kind of fast response method of galvanochemistry firedamp sensor, comprises the steps:

(1) with galvanochemistry firedamp sensor Δ t's former time delay for the sampling period, gather m gas concentration T _i, wherein 0<m<<n, m, n are positive integer, i=1, and 2 ..., m;

(2) m the gas concentration T that step (1) gathers is utilized _iset up discrete grey model DGM (1,1);

(3) utilize the discrete grey model DGM that builds (1,1) in step (2) to carry out forward prediction, obtain gray prediction value wherein ${\tilde{T}}_1=T_1,0<m<<n;$

(4) with the gray prediction value obtained in step (3) for input, actual gas concentration T _ifor exporting, training Gaussian process regression model;

(5) the gray prediction value obtained in step (3) is utilized and the Gaussian process regression model that step (4) is trained, calculate gas concentration C _nbe gas concentration to be asked, firedamp sensor is existed time in concentration when obtaining Δ t, also namely the firedamp sensor response time decreases wherein 0<k<m<n and k is integer.

A kind of fast response method of galvanochemistry firedamp sensor of the present invention, the concrete grammar of described step (2) is:

First, make $Y=\left[\begin{array}{c}{T}^{\left(2\right)}\left(2\right)\\ T^{\left(2\right)\left(3\right)}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ T^{\left(2\right)}\left(m\right)\end{array}\right],B=\left[\begin{array}{cc}{T}^{\left(2\right)}\left(1\right)& 1\\ T^{\left(2\right)}\left(2\right)& 1\\ \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;\\ T^{\left(2\right)}(m-1)& 1\end{array}\right],$ Then ask for $\mathrm{\&beta;}=\left[\begin{array}{c}{\mathrm{\&beta;}}_1\\ {\mathrm{\&beta;}}_2\end{array}\right]{(B^{\&prime;}\&CenterDot;B)}^{-1}\&CenterDot;B^{\&prime;}\&CenterDot;Y,$ Wherein $T^{\left(2\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{T}^{\left(1\right)}\left(j\right),T^{\left(1\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}T\left(j\right),i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m,$ The transposed matrix that B ' is B;

Then, setting up discrete grey model DGM (1,1) is: $\left\{\begin{array}{c}\tilde{T}\left(i\right)={{\mathrm{\&beta;}}_1}^{i-2}[({\mathrm{\&beta;}}_1-1)T\left(1\right)+{\mathrm{\&beta;}}_2],i=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,m\\ \tilde{T}\left(i\right)=T\left(1\right),i=1\end{array}\right\}.$

A kind of fast response method of galvanochemistry firedamp sensor of the present invention, in described step (3), the concrete steps of Gaussian process regression model training are:

(3.1) the covariance function K (x that Gaussian process returns is constructed ⁱ, x ^j) be a square index covariance function K _sE(x ⁱ, x ^j), Rational Quadratic covariance function K _rQwith Matern covariance function K _mthe quadratic sum of three, that is:

$K_{\mathrm{SE}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;e^{-\frac{1}{2}{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}+{\mathrm{\&sigma;}}_n^2{\mathrm{\&delta;}}^{\mathrm{ij}}$

$K_{\mathrm{RQ}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;{[1+\frac{{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}{2\mathrm{\&alpha;}}]}^{-\mathrm{\&alpha;}}$

$K_M(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;[1+\sqrt{3M}(x^i-x^j)\&CenterDot;e^{\sqrt{3M}(x^i-x^j)}]$

K(x ⁱ,x ^j)=[K _SE(x ⁱ,xj)] ²+[K _RQ(x ⁱ,x ^j)] ²+[K _M(x ⁱ,x ^j)] ²

Wherein, M=diag (l ^-2) be the symmetric matrix of hyper parameter, l is relevance measure hyper parameter, σ _n, δ ^ij, α is undetermined parameter, (x ⁱ-x ^j) ^'for (x ⁱ-x ^j) transposed matrix, x ⁱ, x ^jfor input amendment, i ≠ j;

(3.2) the gray prediction value obtained from step (3) in choose successively k value for input, with actual gas density for export, form (m-k) individual input and output training sample pair, that is: input amendment is output sample is T _i+k, i=1,2 ..., m-k;

(3.3) utilize (m-k) in step (3.2) individual input and output training sample pair, it is the single model exported of k input that training obtains Gaussian process regression model, wherein 0<k<m and k is integer.

A kind of fast response method of galvanochemistry firedamp sensor of the present invention, described step (5) obtain gas concentration C to be asked _nconcrete grammar for: with the gray prediction value obtained in step (3) for the input of the Gaussian process regression model that step (4) is trained, then the output of Gaussian process regression model is gas concentration C to be asked _n.

A kind of fast response method of a kind of galvanochemistry firedamp sensor of the present invention, its specific implementation process calculates realization by writing software program in the data processing chips such as single-chip microcomputer, and without the need to changing the component of firedamp sensor own and structure thereof.

After adopting such scheme, the present invention is passed through the some gas concentration T in transient by Quick Acquisition gas density ₁, T ₂..., T _mits T.T. used of sampling is far smaller than settling time, thus final required gas density actual value can be obtained in transient process, and the gas density actual value needed for just obtaining during without the need to reaching real stable state, effectively can solve the latency issue of firedamp sensor.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of a kind of fast response method of galvanochemistry firedamp sensor.

Embodiment

Below with reference to accompanying drawing, technical scheme of the present invention is described in detail.

The invention provides a kind of fast response method of galvanochemistry firedamp sensor, its general thought is the some gas concentrations be passed through by Quick Acquisition gas density in transient, adopt based on DGM-GPR(Discrete Grey Model-GaussianProcesses Regression) prediction algorithm obtain gas density in advance, and then gas be passed through transient do not reach stable state time, to obtain gas concentration during stable state in advance, high degree reduces the response time of firedamp sensor.

Gray prediction based on gray theory, process small sample, poor information, uncertain problem have unique advantage.Discrete grey model DGM (1,1), by traditional gray model GM (1,1) precision, improves the stability of prediction to a certain extent.Gaussian process (Gaussian Processes, GP) as a kind of new machine learning techniques, it is a kind of kernel-based method, first GP carrys out the priori function of Modling model with the form of probability distribution, then under Bayesian frame, realize by the conversion of priori function to posteriority function, and can calculate " hyper parameter " of kernel function.Gaussian process regression algorithm has good adaptability to challenges such as process small sample, non-linear, high dimensions, and the computing time exporting aim parameter predicted value reduces, and may be used for the quick calculating of firedamp sensor concentration value.

Below, by composition graphs 1, step of the present invention is described in detail.A kind of fast response method of galvanochemistry firedamp sensor of the present invention, comprises the following steps:

(1) with galvanochemistry firedamp sensor Δ t's former time delay for the sampling period, gather m gas concentration T _i, wherein 0<m<<n, m, n are positive integer, i=1, and 2 ..., m;

(2) m the gas concentration T that step (1) gathers is utilized _iset up discrete grey model DGM (1,1).

First, make $Y=\left[\begin{array}{c}{T}^{\left(2\right)}\left(2\right)\\ T^{\left(2\right)\left(3\right)}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ T^{\left(2\right)}\left(m\right)\end{array}\right],B=\left[\begin{array}{cc}{T}^{\left(2\right)}\left(1\right)& 1\\ T^{\left(2\right)}\left(2\right)& 1\\ \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;\\ T^{\left(2\right)}(m-1)& 1\end{array}\right],$ Then ask for $\mathrm{\&beta;}=\left[\begin{array}{c}{\mathrm{\&beta;}}_1\\ {\mathrm{\&beta;}}_2\end{array}\right]{(B^{\&prime;}\&CenterDot;B)}^{-1}\&CenterDot;B^{\&prime;}\&CenterDot;Y,$ Wherein $T^{\left(2\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{T}^{\left(1\right)}\left(j\right),T^{\left(1\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}T\left(j\right),i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m,$ The transposed matrix that B ' is B;

Then, setting up discrete grey model DGM (1,1) is: $\left\{\begin{array}{c}\tilde{T}\left(i\right)={{\mathrm{\&beta;}}_1}^{i-2}[({\mathrm{\&beta;}}_1-1)T\left(1\right)+{\mathrm{\&beta;}}_2],i=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,m\\ \tilde{T}\left(i\right)=T\left(1\right),i=1\end{array}\right\};$

Finally, according to built discrete grey model and gas concentration T (1), each moment gray prediction value can be asked for.

(3) utilize the discrete grey model DGM that builds (1,1) in step (2), carry out forward prediction, obtain gray prediction value wherein ${\tilde{T}}_1=T_1,0<m<<n.$

(4) with the gray prediction value obtained in step (3) for input, actual gas concentration T _ifor exporting, training Gaussian process regression model, concrete steps are:

(4.1) the covariance function K (x that Gaussian process returns is constructed ⁱ, x ^j) be a square index covariance function K _sE(x ⁱ, x ^j), Rational Quadratic covariance function K _rQwith the quadratic sum of Matern covariance function KM three, that is:

$K_{\mathrm{SE}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;e^{-\frac{1}{2}{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}+{\mathrm{\&sigma;}}_n^2{\mathrm{\&delta;}}^{\mathrm{ij}}$

$K_{\mathrm{RQ}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;{[1+\frac{{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}{2\mathrm{\&alpha;}}]}^{-\mathrm{\&alpha;}}$

$K_M(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;[1+\sqrt{3M}(x^i-x^j)\&CenterDot;e^{\sqrt{3M}(x^i-x^j)}]$

K(x ⁱ,x ^j)=[K _SE(x ⁱ,x ^j)] ²+[K _RQ(x ⁱ,x ^j)] ²+[K _M(x ⁱ,x ^j)] ²

Wherein, M=diag (l ^-2) be the symmetric matrix of hyper parameter, l is relevance measure hyper parameter, , σ _n, δ ^ij, α is undetermined parameter, (x ⁱ-x ^j) ^￠for (x ⁱ-x ^j) transposed matrix, x ⁱ, x ^jfor input amendment, i ≠ j;

Each parameter (l, , σ _n, δ ^ij, α) value is initialized as random value, adopts method of conjugate gradient, obtains optimized parameter by the maximization of the log-likelihood function (formula (1)) to training sample.Method of conjugate gradient is prior art, does not repeat them here.

$L=-\frac{1}{2}{y}^{\&prime;}{K}^{-1}y--\frac{1}{2}\lg \left|K\right|-\frac{N}{2}\lg \left(2\mathrm{\&pi;}\right)$ Formula (1)

In formula (1), y is that Gaussian process regression model exports data variable, and K is set covariance function, and N is number of training, the transposition that y ' is y.

(4.2) the gray prediction value obtained from step (3) in choose successively k value for input, with actual gas density for export, form (m-k) individual input and output training sample pair, that is: input amendment is output sample is T _i+k, i=1,2 ..., m-k, utilizes this (m-k) individual input and output training sample pair, and it is the single model exported of k input that training obtains Gaussian process regression model, wherein 0<k<m and k is integer.

(5) the gray prediction value obtained from step (3) in choose as the input of the Gaussian process regression model of training for step (4), then the output of Gaussian process regression model is gas concentration C to be asked _n, firedamp sensor is existed time in obtain Δ _ttime concentration, namely also the firedamp sensor response time decreases wherein 0<k<m<n and k is integer.

A kind of fast response method of galvanochemistry firedamp sensor of the present invention, its specific implementation process calculates realization by writing software program in the data processing chips such as single-chip microcomputer, and without the need to changing the component of firedamp sensor own and structure thereof.

Above embodiment is only and technological thought of the present invention is described, can not limit protection scope of the present invention with this, and every technological thought proposed according to the present invention, any change that technical scheme basis is done, all falls within scope.

Claims (3)

1. a kind of fast response method of galvanochemistry firedamp sensor, is characterized in that comprising the following steps:

(1) with galvanochemistry firedamp sensor Δ t's former time delay for the sampling period, gather m gas concentration T _i, wherein 0 < m < < n, m, n are positive integer, i=1, and 2 ..., m;

(2) m the gas concentration T that step (1) gathers is utilized _isetting up discrete grey model DGM (1,1) is $\left\{\begin{array}{c}\tilde{T}\left(i\right){{\mathrm{\&beta;}}_1}^{i-2}[({\mathrm{\&beta;}}_1-1)T\left(1\right)+{\mathrm{\&beta;}}_2],i=\mathrm{2,3},...,m\\ \tilde{T}\left(i\right)=T\left(1\right),i=1\end{array}\right\},$ Concrete methods of realizing is:

First, make $Y=\left[\begin{array}{c}{T}^{\left(2\right)}\left(2\right)\\ T^{\left(2\right)}\left(3\right)\\ .\\ .\\ .\\ T^{\left(2\right)}\left(m\right)\end{array}\right],B=\left[\begin{array}{cc}{T}^{\left(2\right)}\left(1\right)& 1\\ T^{\left(2\right)}\left(2\right)& 1\\ .& .\\ .& .\\ .& .\\ T^{\left(2\right)}(m-1)& 1\end{array}\right],$ Then ask for $\mathrm{\&beta;}=\left[\begin{array}{c}{\mathrm{\&beta;}}_1\\ {\mathrm{\&beta;}}_2\end{array}\right]={(B^{\&prime;}\&CenterDot;B)}^{-1}\&CenterDot;B^{\&prime;}\&CenterDot;Y,$ Wherein $T^{\left(2\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{T}^{\left(1\right)}\left(j\right),T^{\left(1\right)}\left(i\right)=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}T\left(j\right),i=\mathrm{1,2},...,m,$ The transposed matrix that B ' is B;

Then, setting up discrete grey model DGM (1,1) is: $\left\{\begin{array}{c}\tilde{T}\left(i\right){{\mathrm{\&beta;}}_1}^{i-2}[({\mathrm{\&beta;}}_1-1)T\left(1\right)+{\mathrm{\&beta;}}_2],i=\mathrm{2,3},...,m\\ \tilde{T}\left(i\right)=T\left(1\right),i=1\end{array}\right\};$

(3) utilize the discrete grey model DGM that builds (1,1) in step (2), carry out forward prediction, obtain gray prediction value ${\tilde{T}}_1=T_1,$ 0＜m＜＜n；

(4) with the gray prediction value obtained in step (3) for input, actual gas concentration T _ifor exporting, training Gaussian process regression model GPR (Gaussian Processes Regression), concrete methods of realizing is:

(4.1) the covariance function K (x that Gaussian process returns is constructed ⁱ, x ^j) be a square index covariance function K _sE(x ⁱ, x ^j), Rational Quadratic covariance function K _rQwith Matern covariance function K _mthe quadratic sum of three, that is:

$K_{\mathrm{SE}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;e^{-\frac{1}{2}{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}+{\mathrm{\&sigma;}}_n^2{\mathrm{\&delta;}}^{\mathrm{ij}}$

$K_{\mathrm{RQ}}(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;{[1+\frac{{(x^i-x^j)}^{\&prime;}\&CenterDot;M(x^i-x^j)}{2\mathrm{\&alpha;}}]}^{-\mathrm{\&alpha;}}$

$K_M(x^i,x^j)={\mathrm{\&sigma;}}_f^2\&CenterDot;[1+\sqrt{3M}(x^i-x^j)\&CenterDot;e^{\sqrt{3M}(x^i-x^j)}]$

K(x ⁱ,x ^j)＝[K _SE(x ⁱ,x ^j)] ²+[K _RQ(x ⁱ,x ^j)] ²+[K _M(x ⁱ,x ^j)] ²

Wherein, M=diag (l ^-2) be the symmetric matrix of hyper parameter, l is relevance measure hyper parameter, σ n, δ ^ij, α is undetermined parameter, (x ⁱ-x ^j) ' be (x ⁱ-x ^j) transposed matrix, x ⁱ, x ^jfor input amendment, i ≠ j;

(4.2) the gray prediction value obtained from step (3) in choose successively k value for input, with actual gas density for export, form (m-k) individual input and output training sample pair, that is:

Input amendment is: $[{\tilde{T}}_i,{\tilde{T}}_{i+1},...,{\tilde{T}}_{i+k-1}],i=\mathrm{1,2},...,m-k,$

Output sample is: T _i+k,

(4.3) (m-k) constructed by step (4.2) individual input and output training sample pair is utilized, it is the single model exported of k input that training obtains Gaussian process regression model, wherein 0 < k < m and k is integer;

(5) the gray prediction value obtained in step (3) is utilized and the Gaussian process regression model that step (4) is trained, calculate gas concentration C _nbe gas concentration to be asked, firedamp sensor is existed time in concentration when obtaining Δ t, also namely the firedamp sensor response time decreases wherein 0 < k < m < n and k is integer.

2. a kind of fast response method of galvanochemistry firedamp sensor as claimed in claim 1, it is characterized in that, the concrete steps of described step (5) are:

With the gray prediction value obtained in step described in claim 1 (3) for the input of the Gaussian process regression model that step (4) is trained, then the output of Gaussian process regression model is gas concentration C to be asked _n, firedamp sensor is existed time in concentration when obtaining Δ t, also namely the firedamp sensor response time decreases $\frac{n-m}{n}\&CenterDot;\mathrm{\&Delta;t}.$

3. a kind of fast response method of galvanochemistry firedamp sensor as claimed in claim 1, it is characterized in that, its specific implementation process calculates realization by writing software program in the data processing chips such as single-chip microcomputer, and without the need to changing the component of firedamp sensor own and structure thereof.