CN106845447A

CN106845447A - A kind of face gas concentration prediction method for early warning

Info

Publication number: CN106845447A
Application number: CN201710088003.9A
Authority: CN
Inventors: 王苏
Original assignee: Liaoning Technical University
Current assignee: Liaoning Technical University
Priority date: 2017-02-19
Filing date: 2017-02-19
Publication date: 2017-06-13

Abstract

The invention discloses a kind of face gas concentration prediction method for early warning, it is characterized in that, in order to carry out coal mine work area gas monitor concentration prediction early warning, first gas monitor data are carried out with the pretreatment such as abnormal data replacement, Completing Missing Values, de-noising, the integrity of gas density original input data is realized；Then chaotic property identification and phase space reconfiguration are carried out using the time series of the original gas emission of Lyapunov exponent pairs, gas emission chaotic prediction Mathematical Modeling is set up with this；Predicted the outcome according to gas density and its forecast interval carries out threshold value of warning calculating, and its corresponding advanced warning grade is divided, complete early warning information；Can be used for face gas concentration prediction and early warning work.

Description

Working face gas concentration prediction early warning method

Technical Field

The invention relates to mining safety, in particular to working face gas concentration prediction and early warning work.

Background

With the development of the technology, the alarm system with real-time monitoring is already implemented in most mines in coal mines in China. However, the existing monitoring system is mainly limited to local monitoring and management in the downhole, and effective processing and specific analysis of monitoring data aiming at realizing a prediction and early warning function are lacked. According to the actual conditions and requirements of a mine, by taking the rule of mine gas monitoring data as an analysis means, the abnormal conditions in the gas emission process are specifically analyzed, the technical research of the prediction and early warning analysis of the mine gas concentration is carried out, real and effective prediction results are given, the gas early warning of real-time prediction is realized, the early warning level is determined, the early warning information is issued, the gas early warning is transferred to the front after the coal mine ventilation safety management work is carried out, and the important practical significance is realized for the enhancement of the coal mine ventilation safety management work and the improvement of the gas disaster prevention and control level.

1 mine gas monitoring data processing

Due to the special and complex production environment of the underground coal mine and the limitation of the monitoring system, the monitoring data has the possibility of data abnormity, data loss, noise and the like. Therefore, the gas monitoring data collected by the monitoring system can show complex and nonlinear characteristics. The prediction precision can be improved as much as possible through data processing processes such as abnormal data replacement, missing data completion, noise elimination and the like.

1.1 gas monitoring anomaly data handling

Abnormal values may occur in the gas concentration monitoring data at a certain time period or point. The maximum value in the abnormal data is probably caused by catastrophe and has small occurrence probability, so that the maximum value cannot be simply removed and is processed according to the occurrence frequency of the maximum value; the data with zero in the abnormal data may be because the detection signal is disturbed and cannot be simply eliminated.

Let the time series of gas concentration composed of the original gas monitoring data be { x }_tAnd t is 1,2,.., N }, and N is the sequence length. When at a certain time t ═ t₀When an abnormal value occurs and the abnormal value is zero or a high gas concentration value with a lower frequency, the first N in the sequence of the point can be used₀The number is calculated by cubic spline interpolation, and the corresponding value in the interpolation sequence value is usedTo make a substitution; and its substitute value can also be obtained by changing t ═ t₀First N of time₀And carrying out cubic spline interpolation optimization calculation on the data again. And the actual monitoring data is selected as the basis during interpolation calculation, so that the transmission of calculation errors is avoided.

1.2 gas monitoring data loss handling

The method is characterized in that the condition of data loss possibly occurs in data monitored by a coal mine monitoring system in real time at a certain time period or time point, and the method is suitable for short-term prediction calculation under the condition that a series of data is in a quadratic parabola growth trend for a long time according to the characteristics of an exponential smoothing algorithm, so that the method has the function of resisting and weakening abnormal data. Therefore, the data are complemented by adopting a time series 3-degree exponential smoothing method, namely a Brown quadratic polynomial exponential smoothing method.

1.3 gas monitoring data noise elimination processing

Due to the combined action of uncontrollable factors in human and environment aspects in the processes of data acquisition, transmission, storage, processing and the like, the signal contains noise influence. The existence of the noise data not only affects the accuracy and reliability of the data, but also enables the sequence to show the dispersity. Therefore, the acquired data needs to be subjected to wavelet denoising processing.

(1) The gas concentration monitoring data is subjected to wavelet decomposition, and after abnormal value processing and missing data supplementing processing are carried out on the data, the monitoring data still has large fluctuation, and the distribution of the maximum values is not centralized. Wavelet de-noising processing is carried out, wavelet decomposition processing is carried out on the wavelet de-noising processing, and scale coefficients c are obtained respectively_j,kSum wavelet coefficient d_j,k. The maximum number of decomposition layers required for noise cancellation cannot exceed 5, since the data is still true and valid here.

(2) And (3) threshold processing of wavelet decomposition coefficients, namely determining a threshold by adopting an unbiased risk estimation criterion method with self-adaptive characteristics on the basis of ensuring the authenticity of data for gas concentration monitoring data, namely solving a corresponding risk value for a given threshold lambda and selecting the minimum risk valueThe value is obtained. Sorting the square values of the wavelet coefficients obtained by the previous decomposition from small to large to form a vector W (W (k)), k being 1,2_kAnd defining its risk value as:

(3) reconstructing inverse wavelet transform, and obtaining wavelet decomposition coefficient by the above calculationAnd a scale factor c_j,kAnd performing wavelet inverse transformation, wherein the sequence obtained by reconstruction is the gas concentration time sequence obtained after wavelet denoising.

2 working face gas concentration time series correlation analysis

2.1 identification of chaotic time series

And judging the chaos of the time sequence by utilizing the Lyapunov exponent, wherein if the maximum Lyapunov exponent is more than 0, the dynamic system is chaotic. The Lyapunov index is calculated by a Wolf method: assuming an embedding dimension m and a time delay τ, the reconstruction phase space is formulated for the time series x (1), x (2),.. the x (t),. the reconstructed phase space is:

X(t)＝{x(t),x(t+t),...,x[t+(m-1)t]},t＝1,2,...,M (2)

noting the initial time as t₀The current time is t_iThe end time is t_MM ═ N- (M +1) τ, and the sequence end point is denoted N. Let the initial point be X (t)₀) With the nearest neighbor point X₀(t₀) Is a distance L₀Passing through t_iTime of day, the distance of which exceeds a predetermined threshold (> 0), and L₀＝|X(t_l)-X(t₀)|>. At X (t)_l) Search for another point X in the field_l(t_l) So that L is_l＝|X(t_l)-X_l(t_l)|<Tracking its evolution process until x (t) reaches the end of the time sequence N, the total number of iterationsIs t_M-t₀The maximum Lyapunov exponent λ₁Obtaining:

wherein, L'_i＝|X(t_i)-X(t_i-1)|，L_i＝|X(t_i)-X_i(t_i)|，X_i(t_i) Is t_iAt time, in state X (t)_i) As a point in the radius domain.

2.2 chaotic time series phase space reconstruction

Takens proposed the delayed coordinate embedding theorem in 1981, demonstrating that the system phase space can be reconstructed from a single time series. Takens' theorem: m is a d-dimensional manifold, φ: m → M, y: M → R, y has a second continuous derivative, φ (φ, y): M → R^2d ⁺¹，f(φ,y)＝(y(x),y(φ(x)),y(φ²(x)),...,y(φ^2d(x) Phi (phi, y)) is M to R)^2d+1One of (a) is embedded.

From the above theorem, there is an embedded dimensional space that can recover the regular trajectory. Thereby reconstructing a phase space, which can be used to estimate the original system phi. The time delay τ and the embedding dimension m of this embedding dimension are obtained here by computational analysis by the C-C method and the G-P method.

2.3 chaotic time series prediction

The chaos time sequence prediction principle is that after a phase space is reconstructed, an attractor is recovered to fit an adjacent state model. A first-order weighted local area method is mainly applied to construct a gas emission quantity chaotic prediction mathematical model. The first-order weighted local prediction method is that in a phase space, a vector set is found out as a reference set by calculating Euclidean distances between each neighborhood point and X (t), namely: x (t)_i) 1,2, and d_iTo represent a point X (t)_i) And X (t) and d_minIs d_iTo thereby calculate a definition point X (t)_i) The formula of the weight is as follows:

where a is a constant coefficient, where a is 1, the linear fitting form is:

X(t_i+1)＝aR+bX(t_i)(5)

wherein R ═ 1, 1.., 1]^T. Each phase point can be decomposed into:

wherein,represents X (t)_i+1) of the j-th dimension, m being the embedding dimension.

According to the weighted least squares method:

considering equation (7) as a univariate function for a and b, and simultaneously obtaining the partial derivatives for a and b, respectively, there are:

solving an equation set to obtain a and b, carrying out fitting validity analysis on the a and b to obtain evolution phase point prediction, and extracting the final one-dimensional component to obtain a predicted value.

Disclosure of Invention

1. A working face gas concentration prediction early warning method is characterized in that in order to carry out gas monitoring concentration prediction early warning on a coal mine working face, preprocessing such as abnormal data substitution, missing data completion and noise elimination is carried out on gas monitoring data, and therefore the integrity of original input data of the gas concentration is achieved; then, performing chaos recognition and phase space reconstruction on the time sequence of the original gas emission quantity by using a Lyapunov index so as to establish a gas emission quantity chaos prediction mathematical model; calculating an early warning threshold according to the gas concentration prediction result and the prediction interval thereof, and dividing early warning grades corresponding to the early warning threshold to finish early warning information; the method can be used for working face gas concentration prediction and early warning work.

2. The working face gas concentration prediction and early warning method according to claim 1, characterized in that the basic index determination of the early warning of the gas emission quantity at the monitoring point is carried out through the mean value mu of the monitoring data of the gas concentration_xSum variance σ_xCalculating and determining the time t when the predicted value of the gas emission is continuously larger_h(ii) a According to different situations, the confidence intervals corresponding to different confidence levels can be divided into the following three types: confidence intervals at 95% confidence level [ mu-1.96 sigma,. mu. +1.96 sigma ]]Confidence interval at 85% confidence level [ mu-1.44 sigma,. mu. +1.44 sigma ]]And confidence intervals at 68.3% confidence level [ mu-sigma, [ mu ] + sigma ]]。

3. The working face gas concentration prediction and early warning method as claimed in claim 1, wherein the early warning I stage comprises: if the gas concentration is predicted and the upper bound x of the prediction interval_*+_*≤μ_x+σ_xJudging that the predicted value is at a confidence level of 68.3%, and judging that the predicted value is normal, and not carrying out early warning; at x_*<w_sOn the premise of (1), when x_*+_*∈[μ_x+σ_x,μ_x+1.44σ_x]If so, judging that the predicted value is under the 85% confidence level, and considering the predicted value as a normal condition without warning; otherwise, the gas concentration is considered to be larger, if the predicted value is larger continuously within 1 hour and the time is t_hWhen t is_h>When the time is 30min, the abnormal condition is determined, the grade is determined to be early warning I grade, and warning time t is set_h≤t_w≤1h，t_wIndicating an alert time.

4. The working face gas concentration prediction and early warning method as claimed in claim 1, wherein the early warning level II comprises: when x is_*+₂∈[μ_x+1.44σ_x,μ_x+1.96σ_x]Judging the confidence level between the predicted value of 85 percent and 95 percent, considering the normal condition, and not carrying out early warning; when x is_*+₂>μ_x+1.96σ_xIf the predicted value is continuously larger and the time is t within 1 continuous hour when the alarm threshold value is not reached_hWhen t is_h>And when the time is 30min, judging the confidence level between 85% and 95% of the predicted value, determining that the predicted value has a continuous greater trend, determining that the predicted value is abnormal, determining the grade as early warning II grade, and setting a warning time interval as follows: t is t_h≤t_w≤1h。

5. The working face gas concentration prediction and early warning method as claimed in claim 1, wherein early warning level III: when x is_*+₂>μ_x+1.96σ_xWhen the alarm threshold value is not reached, the alarm continues to be larger within 2 continuous hours for t_hIf t is_h>And (5) 60min, indicating that the predicted value does not exceed the alarm concentration and is not at the 95% confidence level, and the predicted value has a continuous greater trend, determining the abnormal condition, determining the grade as early warning grade III, and setting an alarm time interval as follows: t is t_h≤t_wLess than or equal to 2 h; if it is 30min<t_h<And (5) determining the grade as early warning II grade after 60 min.

Drawings

Time sequence partial data segment of original data of 1# air inlet of working face of FIG. 12101

FIG. 2 time-series enlarged view of abnormal value processed

FIG. 3 time series after wavelet de-noising

1# air-return tunnel entrance gas emission early warning information display of figure 42101 working surface

Detailed Description

Taking a 1# air return tunnel entrance on a working face of a frequently-village coal mine 2101 as an example, 15 days from 6 months 1 day in 2015 to 6 months 15 days in 2015 are respectively monitored by gas monitoring points for 4320 data, the time length is 21600min, the average time interval is 5min, the maximum value of the gas concentration is 0.311%, the minimum value is 0.216%, the data are used as original data, and a data preprocessing method is adopted for processing the data. The time series partial data thereof is shown in fig. 1.

An abnormal value processing was performed on the gas concentration time series by a cubic spline interpolation method, and an enlarged view of the processed time series is shown in fig. 2.

The original data of the gas concentration monitoring is complete, the completion processing is not needed, but the wavelet denoising processing is needed to be carried out on the sequence, and the obtained gas concentration time sequence is shown in figure 3. On the basis of keeping the average trend of the original time series, the data curve becomes smooth, and the time series instantaneous features are easy to extract.

The time delay τ is calculated using the C-C method. With the aid of Matlab software, a time delay of 10 was obtained for the inlet of the 1# air inlet at the working face 2101. The embedding dimension m is obtained by adopting a G-P method, and the embedding dimension m of the 1# air inlet of the working surface 2101 obtained by running a self-programming method is 20. And (4) reconstructing a phase space by using the time delay tau and the embedding dimension m obtained by the calculation and analysis. And (4) calculating the Lyapunov index lambda of the 1# air inlet test point to be 0.0406 according to the small data quantity algorithm step. The maximum time that can be predicted is within 25 moments.

On the basis of a predicted value calculated by using a prediction method for the gas concentration monitoring data of the No. 1 air-return tunnel entrance of the 2101 working face, the confidence level of the predicted value of the gas emission quantity of the 2101 working face at 95 percent is calculatedThe average value of_x0.399344, variance σ_xWhen 0.009989, the prediction interval is: [ x ] of_*-_*,x_*+_*]Namely: [0.389355,0.409334]. The obtained 2101 working face # 1 air-return tunnel entrance gas concentration early warning information is displayed as shown in fig. 4.

From fig. 4, it can be known that the predicted value of the gas emission quantity of the 1# air inlet of the 2101 working face is compared with the actually measured value in a time zone of 5 hours, and the predicted value of the predicted point is within the early warning interval range, so the predicted value is consistent with the actual monitoring condition, and the prediction method is feasible.

Claims

2. The working face gas concentration prediction and early warning method according to claim 1, characterized in that the basic index determination of the early warning of the gas emission quantity at the monitoring point is carried out through the mean value mu of the monitoring data of the gas concentration_xSum variance σ_xCalculating and determining the time t when the predicted value of the gas emission is continuously larger_h(ii) a According to different situations, corresponding confidence intervals under different confidence levels^[10]The method can be divided into the following three types: confidence intervals at 95% confidence level [ mu-1.96 sigma,. mu. +1.96 sigma ]]Confidence interval at 85% confidence level [ mu-1.44 sigma,. mu. +1.44 sigma ]]And confidence intervals at 68.3% confidence level [ mu-sigma, [ mu ] + sigma ]]。