CN106296435A - A kind of mine gas monitoring disorder data recognition method - Google Patents

Abstract

The invention discloses a kind of mine gas monitoring disorder data recognition method, consider the factor of production impact on Gas, Real-time Monitoring Data sample is reconstructed, arrange initial centroid vector to cluster, the differentiation sample in cluster is analyzed, if there is sudden change situation, and outside historical data 95% confidence interval, it is then small probability event, determines that monitoring is abnormal, otherwise for normally detecting data.This method can effectively identify that coal mine gas monitors abnormal data, from down-hole diverse location Gas, flow and accumulation feature, consider ventilation options, the factor such as the flowing law of fluid and Gas Accumulation in Upper-Corner by Pulsed source in ventilation network, based on Historical Monitoring data statistic analysis, by with Safety monitoring system on-line analysis, screening gas monitor abnormal data, solving spurious signal in gas monitor information affects the highest problem of Monitoring Data process computational accuracy and monitoring wrong report problem.

Description

A kind of mine gas monitoring disorder data recognition method

[technical field]

The invention belongs to coal mine safety monitoring and monitoring technical field, be specifically related to a kind of mine gas abnormal data monitoring Recognition methods, big data based on Safety monitoring system real time data process and safe early warning application.

[background technology]

Gas Disaster is one of disaster maximum to the shaft production extent of injury, and Safety of Coal Mine Production, ore deposit in serious threat Well gas monitor is the important means of prevention Gas Disaster, and underground coal mine primary location and region are the pass key mappings of gas monitor Putting, Monitoring Data derives from detecting element methane transducer, and Monitoring Data is transferred to monitor master by sensor by underground substation Machine, monitoring host computer can external data capture program, Real-time Monitoring Data is applied to security early-warning analysis, i.e. by a large amount of watts This Monitoring Data processes and extracts Gas Emission Law, thus can effectively understand the gas that distinguished and admirable middle gas density implied and gush The trend go out, flow, gathered, and then gas density pre-alarm can be realized, belong to big data and process category, for promoting safe critical point Reach has important realistic meaning.

For methane Concentration Measurement, the reliability of methane transducer itself and accuracy rely on quality in site access with special The adjustment of industry technical staff controls, during day-to-day operation, due to rings such as conditions down-hole complexity, temperature, water vapour, mine dusts Border factor can affect the accuracy of methane transducer；Electromagnetic interference in signals transmission and communication line are by impacting with high pressure etc. Factor can cause spurious signal to produce, and therefore transmits to the Monitoring Data of monitoring host computer and inevitably there is abnormal number According to.

Identification with regard to gas monitor abnormal data at present there is no systematic Study achievement and submits scene application to, and based on gas Data that Monitoring Data is carried out process and application must be based on data more accurate, real, thus Study on Coal mine Carapax et Plastrum Testudinis The identification of alkane Sensor monitoring abnormal data with seem most important.

[summary of the invention]

The technical problem to be solved is for above-mentioned deficiency of the prior art, it is provided that a kind of mine gas Monitoring disorder data recognition method, it is possible to effectively identify coal mine gas monitoring abnormal data.

The present invention is by the following technical solutions: a kind of mine gas monitoring disorder data recognition method, described method considers The factor of production impact on Gas, is reconstructed Real-time Monitoring Data sample, arranges initial centroid vector and clusters, Differentiation sample in cluster being analyzed, if there is sudden change situation, and outside historical data 95% confidence interval, being then little Probability event, is defined as monitoring abnormal, is otherwise normal Monitoring Data.

Further, comprise the following steps:

S1, diverse location monitoring point, down-hole wind speed, gas concentration monitoring data sequence are converted into gas flow sequence, shape Become sequence sets D, and calculate Historical Monitoring data mean value μ and standard deviation sigma respectively；

S2, the Monitoring Data in sequence sets D in step S1 is carried out m dimension sample reconstruct, and set 4 classifications, determine each The initial centroid vector of classification；

S3, setting based on step S2, (i p), determines that error is put down to calculate the Euclidean distance d of each sample and each barycenter All samples are divided to 4 classifications by side and E (τ)；

The average centroid distance μ of sample in each classification divided in S4, respectively calculation procedure S3_pAnd standard deviation sigma_p, determine Sample undetermined, calculate sample sequence difference and with historical data statistical nature carry out contrast judge identification sample undetermined the most different Often, for methane prediction.

Further, in described step S1, each monitoring point monitoring situation is converted into gas flow data sequence Q_tSet up Variables set D, to variables set D={Q_1t,Q_2t,Q_3t,Q_4t,Q_5tCarry out disorder data recognition, wherein, t=1,2 ..., Q_t=x_t×v × S, x_tFor monitoring point gas monitor data sequence, v is wind speed, and S is that drift section amasss, Q_1tFor dynamo-electric, material chamber monitoring number According to sequence, Q_2tFor intake Monitoring Data sequence, Q_3tFor return aircourse Monitoring Data sequence, Q_4tMonitor for stope Data sequence, Q_5tFor upper corner Monitoring Data sequence.

Further, in step S2, take the N number of Monitoring Data sample in described sequence sets D, carry out m dimension sample reconstruct, if Fixed 4 classifications K₁,K₂,K₃,K₄, according to the maenvalue point after the reconstruct of timing node statistical history Monitoring Data, as first prothyl Heart vector, calculate each sample and each barycenter Euclidean distance d (i, p):

$d(i,p)=\sqrt{{(x_{ik}-x_{pk})}^2}---\left(1\right)$

Wherein, p=1,2,3,4, i=1,2 ..., N, k=1,2 ..., m, x_tFor monitoring point gas monitor data sequence.

Further, according to being calculatedTime p classify x_i∈K_pIf being divided in certain classification Number of samples is N_p, all kinds of barycenter samples are c_p, then error sum of squares E (τ) particularly as follows:

$E\left(\mathrm{\τ}\right)=\underset{p=1}{\overset{4}{\Σ}}\underset{j=1}{\overset{N_p}{\Σ}}{(x_{ik}-c_p)}^2---\left(2\right).$

Further, it is iterated described error sum of squares E (τ) calculating, if E (τ) ＜ E (τ-1), then with Different categories of samples Mean vector is that barycenter recalculates sample distance, has otherwise clustered.

Average centroid distance μ further, after described cluster completes, in each cluster of calculating_pParticularly as follows:

${\mathrm{\μ}}_p=\frac{1}{N_p}\underset{j=1}{\overset{N_p}{\Σ}}d(j,p)---\left(3\right)$

Wherein, N_pNumber of samples in clustering for pth, (j p) is the jth sample distance to barycenter to d.

Further, primary election sample and centroid distance are more than described average centroid distance μ_pSample, in the cluster of calculating All samples and the standard deviation sigma of average centroid distance_pParticularly as follows:

${\mathrm{\σ}}_p=\sqrt{\frac{1}{N_p}\underset{j=1}{\overset{N_p}{\Σ}}{\left(d\right(j,p)-{\mathrm{\μ}}_p)}^2}---\left(4\right)$

Wherein, μ_pAverage centroid distance in clustering for pth.

Further, in described step S4, for sample belonging to real time data, estimate with normal distribution 95% confidence level Meter, if d is (j, p)-μ_p＞ 1.96 σ_p, it is determined that for abnormal data sample undetermined.

Further, the difference Δ x of adjacent two elements of described exceptional sample undetermined is calculated_jIf this vector existing and being more than The component of historical data sequence maximum difference Δ ax, and actual gas monitor value x_jk＞ μ+1.96 σ or x_jk＜ μ-1.96 σ, then know Wei abnormal data.

Compared with prior art, the present invention at least has the advantages that

One mine gas of the present invention monitoring disorder data recognition method, it is considered to the factor of production impact on Gas, Use sample reconstruct and cluster analysis thereof can be prevented effectively from and actual gas effusion intensity causes distinguished and admirable middle gas density higher (including paroxysmal high gas emission situation) is mistaken for the situation that monitoring is abnormal.

Further, the present invention is by being converted into gas flow data sequence Q by each monitoring point gas monitor data_tSet up Variables set D, it is possible to identify the gas monitor exceptional value occurred due to reasons such as environmental factors, electromagnetic interference, transmission faults is right The abnormal data of this method identification is taked suitably to process, i.e. available close to real gas monitor data sequence, it is to avoid monitoring The existence of exceptional value affects the rule of gas monitor data sequence and extracts and subsequent treatment and analysis application, the method is encapsulated into Monitoring and controlling system can be avoided due to the abnormal wrong report phenomenon caused of monitoring.

Further, this method can also be connected with outside monitoring host computer, by Safety monitoring system on-line analysis, Screening gas monitor abnormal data, solving spurious signal in gas monitor information affects Monitoring Data process computational accuracy The highest problem and monitoring wrong report problem.

Below by drawings and Examples, technical scheme is described in further detail.

[accompanying drawing explanation]

Fig. 1 is schematic flow sheet of the present invention；

Fig. 2 is material depot Monitoring Data sequence and outlier identification schematic diagram；

Fig. 3 is work surface Monitoring Data sequence and outlier identification schematic diagram.

[detailed description of the invention]

For effectively identifying coal mine gas monitoring abnormal data, from down-hole diverse location Gas, flow and accumulation Feature is set out, and system carries out down-hole critical positions and region methane transducer monitoring anomaly analysis, considers ventilation options, logical The factor such as the flowing law of fluid and Gas Accumulation in Upper-Corner by Pulsed source in wind network, based on Historical Monitoring data statistic analysis, The Changing Pattern analyzing different monitoring points gas monitor data identifies abnormal data, uses this method can complete and safety monitoring During monitoring system on-line analysis, the examination of gas monitor abnormal data, solve spurious signal impact in gas monitor information Monitoring Data processes the highest problem of computational accuracy and monitoring wrong report problem.

The invention provides a kind of mine gas monitoring disorder data recognition method, divide based on Historical Monitoring data statistics Analysis, the cluster analysis after reconstructing with Real-time Monitoring Data sample determines abnormal data.Sample reconstruct and the meaning of cluster analysis thereof Justice is to consider the factor of production impact on Gas, it is to avoid actual gas effusion intensity is caused distinguished and admirable middle gas density relatively It is abnormal that high (including paroxysmal high gas emission situation) is mistaken for monitoring, if there is gas effusion intensity situation, gas is supervised Survey data to dramatically increase and continue for some time, clusters number is defined as 4 classes and represents 3 different order of classes or grades at school and owing to producing The order of classes or grades at school size hierarchical level on impact the formed gas monitor data of next order of classes or grades at school, thus significantly high methane concentration is supervised Measured value can be divided among same cluster, and produces a differentiation sample undetermined, and the m data in this sample, according to it Rate of change differentiates again, if there is sudden change situation, and outside historical data 95% confidence interval, is small probability event, I.e. it is considered monitoring abnormal, otherwise is gas effusion intensity；The meaning of historical data statistical analysis is to consider due to geology bar The part isometric cycle influences factor change impact on Gas, defines the gas monitor data of small probability.

Refer to shown in Fig. 1, the invention provides a kind of mine gas monitoring disorder data recognition method, including following step Rapid:

S1, by Sequence Transformed for the Monitoring Data of diverse location monitoring point, down-hole wind speed, gas density for gas flow sequence, Formation sequence collection D, and calculate Historical Monitoring data mean value μ and standard deviation sigma respectively；

Wherein, each monitoring point monitoring situation is converted into gas flow data sequence Q_tSet up variables set D, to variables set D ={ Q_1t,Q_2t,Q_3t,Q_4t,Q_5tCarry out disorder data recognition, wherein, t=1,2 ..., Q_t=x_t× v × S, x_tFor monitoring point watt This Monitoring Data sequence, v is wind speed, and S is that drift section amasss, Q_1tFor dynamo-electric, material chamber Monitoring Data sequence, Q_2tFor air intake Tunnel monitor data sequence, Q_3tFor return aircourse Monitoring Data sequence, Q_4tFor stope Monitoring Data sequence, Q_5tFor upper Corner Monitoring Data sequence.Due to down-hole diverse location Gas, flow, gather and be respectively arranged with its feature, monitoring point gas is supervised Survey data sequence x_t, t=1,2 ..., if coming from electromechanics, material chamber, the main methane accumulation situation considering that low wind speed causes, It is translated into gas flow sequence Q_1t, t=1,2 ..., Q_1t=x_t× v × S, v, S are respectively wind speed and drift section amasss；If Come from the important areas such as getting working face and need labor, if entering, monitoring point, return aircourse gas flow Q_2t、Q_3t, actual mining Monitoring point, face mainly considers rib and coal breakage Gas, is translated into sequence of differences Q_4t=Q_4t-Q_2t, upper corner monitoring point Main consideration goaf gas is gushed out, and is converted into sequence of differences Q_5t=Q_4t-Q_5t.With Q_1t,Q_2t,Q_3t,Q_4t,Q_5tCollectively constitute watt This Analysis on monitoring data sequence sets D, uses following methods to carry out disorder data recognition.

S2, the Monitoring Data in sequence sets D in step S1 is carried out m dimension sample reconstruct, and set 4 classifications, determine each The initial centroid vector of classification；

Particularly as follows: for the N number of Monitoring Data sample in D, its reconstruct dimension is m, sets 4 classifications K₁,K₂,K₃,K₄, According to the maenvalue point after timing node 0 point, 8 points, 10 points, 16 statistics Historical Monitoring data reconstruction of 3 months, as just Prothyl Heart vector, calculates the Euclidean distance of each sample and each barycenter

$d(i,p)=\sqrt{{(x_{ik}-x_{pk})}^2}---\left(1\right)$

Wherein, p=1,2,3,4, i=1,2 ..., N, k=1,2 ..., m, x_tFor monitoring point gas monitor data sequence.

S3, setting based on step S2, (i p), determines that error is put down to calculate the Euclidean distance d of each sample and each barycenter All samples are divided to 4 classifications by side and E (τ)；

By being calculatedTime p classify x_i∈K_pIf being divided into the number of samples in each classification is N_p, all kinds of barycenter samples are c_p, calculate error sum of squares

$E\left(\mathrm{\τ}\right)=\underset{p=1}{\overset{4}{\Σ}}\underset{j=1}{\overset{N_p}{\Σ}}{(x_{ik}-c_p)}^2---\left(2\right)$

It is iterated described error sum of squares E (τ) calculating, after iterative computation, if E (τ) ＜ E (τ-1), with respectively Class sample mean vector is that barycenter recalculates sample distance, has otherwise clustered.

Average centroid distance μ in each classification divided in S4, respectively calculation procedure S3_pAnd standard deviation sigma_p, determine undetermined Sample, calculates sample sequence difference and carries out contrast with historical data statistical nature and judge sample undetermined whether exception.

Wherein, for sample belonging to real time data, estimate with normal distribution 95% confidence level, if d is (j, p)-μ_p＞ 1.96σ_p, it is determined that for abnormal data sample undetermined；Calculate the adjacent element difference vector Δ x of described exceptional sample undetermined_jIf, should Vector exists the component more than historical data sequence maximum difference Δ ax, and actual gas monitor value x_jk＞ μ+1.96 σ or x_jk ＜ μ-1.96 σ, then be judged as abnormal data.

The average centroid distance μ of each apoplexy due to endogenous wind is calculated after completing cluster_p, primary election sample and centroid distance more than average barycenter away from From sample, calculate the standard deviation sigma of all samples of apoplexy due to endogenous wind and barycenter average distance_p, estimate with normal distribution 95% confidence level, If d is (j, p)-μ_p＞ 1.96 σ_p, it is determined that this sample is abnormal data sample undetermined.At sample x undetermined_jIn comprise m monitoring number According to, seek sample adjacent element difference vector Δ x_jIf, Δ x_j＞ Δ ax, Δ ax are historical data maximum difference, and there is x_jk＞ μ+ 1.96 σ or x_jk＜ μ-1.96 σ, then can be identified as abnormal data, otherwise, for normal data, wherein μ, σ are respectively Historical Monitoring number According to average and standard deviation, Historical Monitoring data take the data analysis of 1a.For identified monitoring abnormal data carry out interpolation or Smooth replacement i.e. can obtain weakening due to the monitoring exceptional value impact on subsequent calculations precision.

By online with external security monitoring and controlling system, it is possible in real time gas monitor abnormal data is discriminated online Not, solve spurious signal in gas monitor information by computer disposal and affect the problem that Monitoring Data process computational accuracy is the highest With monitoring wrong report problem.

Embodiment

Refer to shown in Fig. 2, for ore deposit, Shaanxi material depot monitoring point gas monitor data sequence, have chosen in March, 2015 Producing 3000 Monitoring Data of class, the monitoring cycle is 30s, uses above method identification monitoring exceptional value therein, monitors different Constant value comprises the situation that the reason such as electromagnetic interference, transmission fault causes monitor value to be zero that is likely to be due to, and instantaneous higher value, this Two class exceptional values deviate considerably from gas monitor data sequence entirety tendency, monitor value be zero situation for monitoring exceptional value be aobvious and Being clear to, the instantaneous higher value in gas monitor value is mutation status, and beyond historical data 95% confidence interval, be considered as little generally Rate situation, and the possibility of abnormal Gas is eliminated through cluster analysis, thus it is judged to thundering observed data；

Refer to shown in Fig. 3, for this ore deposit stope monitoring point gas monitor data sequence, choose same order of classes or grades at school 700 Monitoring Data, use above method identification abnormal data before front 300 data points, at this moment between section gas dense Degree entirety is the highest, but owing to the reasons such as environmental factors, electromagnetic interference, transmission fault cause these monitorings of appearance of monitoring exceptional value Abnormal data deviates considerably from the gas density size overall trend in this time period, shows mutability and beyond historical data 95% confidence interval, is considered as small probability situation, and eliminates the possibility of abnormal Gas through cluster analysis, thus judges For thundering observed data；After section, gas density is on the rise at this moment and undulatory property is relatively strong, part monitor value close to The monitoring exceptional value of first half section, but differentiate that by cluster analysis belonging to gas emission increases caused.By using the method, identify Go out gas monitor exceptional value, and use suitable method to process it, i.e. available close to real gas monitor data sequence, The rule avoiding the existence monitoring exceptional value to affect gas monitor data sequence is extracted and subsequent treatment and analysis application, by the party Method is encapsulated into monitoring and controlling system and can avoid due to the abnormal wrong report phenomenon caused of monitoring.

Above content is only the technological thought that the present invention is described, it is impossible to limit protection scope of the present invention with this, every presses The technological thought proposed according to the present invention, any change done on the basis of technical scheme, each fall within claims of the present invention Protection domain within.

Claims (10)

1. a mine gas monitoring disorder data recognition method, it is characterised in that described method considers that the factor of production is to gas The impact gushed out, is reconstructed Real-time Monitoring Data sample, arranges initial centroid vector and clusters, to the differentiation in cluster Sample is analyzed, if there is sudden change situation, and outside historical data 95% confidence interval, is then small probability event, determines Abnormal for monitoring, it is otherwise normal Monitoring Data.

A kind of mine gas monitoring disorder data recognition method the most according to claim 1, it is characterised in that include following Step:

S1, diverse location monitoring point, down-hole wind speed, gas concentration monitoring data sequence are converted into gas flow sequence, form sequence Row collection D, and calculate Historical Monitoring data mean value μ and standard deviation sigma respectively；

S2, the Monitoring Data in sequence sets D in step S1 is carried out m dimension sample reconstruct, and set 4 classifications, determine of all categories Initial centroid vector；

S3, setting based on step S2, (i p), determines error sum of squares to calculate the Euclidean distance d of each sample and each barycenter All samples are divided to 4 classifications by E (τ)；

The average centroid distance μ of sample in each classification divided in S4, respectively calculation procedure S3_pAnd standard deviation sigma_p, determine undetermined Sample, calculates sample sequence difference and carries out contrast with historical data statistical nature and judge to identify that sample undetermined is whether abnormal, use In methane prediction.

A kind of mine gas monitoring disorder data recognition method the most according to claim 2, it is characterised in that described step In S1, each monitoring point monitoring situation is converted into gas flow data sequence Q_tSet up variables set D, to variables set D={Q_1t,Q_2t, Q_3t,Q_4t,Q_5tCarry out disorder data recognition, wherein, t=1,2 ..., Q_t=x_t× v × S, x_tFor monitoring point gas monitor data Sequence, v is wind speed, and S is that drift section amasss, Q_1tFor dynamo-electric, material chamber Monitoring Data sequence, Q_2tNumber is monitored for intake According to sequence, Q_3tFor return aircourse Monitoring Data sequence, Q_4tFor stope Monitoring Data sequence, Q_5tNumber is monitored for upper corner According to sequence.

A kind of mine gas monitoring disorder data recognition method the most according to claim 3, it is characterised in that step S2 In, take the N number of Monitoring Data sample in described sequence sets D, carry out m dimension sample reconstruct, set 4 classifications K₁,K₂,K₃,K₄, press According to timing node statistical history Monitoring Data reconstruct after maenvalue point, as initial centroid vector, calculate each sample with The Euclidean distance d of each barycenter (i, p):

$d(i,p)=\sqrt{{(x_{ik}-x_{pk})}^2}---\left(1\right)$

Wherein, p=1,2,3,4, i=1,2 ..., N, k=1,2 ..., m, x_tFor monitoring point gas monitor data sequence.

A kind of mine gas monitoring disorder data recognition method the most according to claim 4, it is characterised in that according to calculating ObtainTime p classify x_i∈K_pIf being divided into the number of samples in certain classification is N_p, all kinds of barycenter samples For c_p, then error sum of squares E (τ) particularly as follows:

$E\left(\mathrm{\τ}\right)=\underset{p=1}{\overset{4}{\Σ}}\underset{j=1}{\overset{N_p}{\Σ}}{(x_{ik}-c_p)}^2---\left(2\right).$

A kind of mine gas monitoring disorder data recognition method the most according to claim 5, it is characterised in that to described mistake Difference quadratic sum E (τ) is iterated calculating, if E (τ) ＜ E (τ-1), then recalculates sample with Different categories of samples mean vector for barycenter This distance, has otherwise clustered.

A kind of mine gas monitoring disorder data recognition method the most according to claim 6, it is characterised in that described cluster Average centroid distance μ after completing, in each cluster of calculating_pParticularly as follows:

${\mathrm{\μ}}_p=\frac{1}{N_p}\underset{j=1}{\overset{N_p}{\Σ}}d(j,p)---\left(3\right)$

Wherein, N_pNumber of samples in clustering for pth, (j p) is the jth sample distance to barycenter to d.

A kind of mine gas monitoring disorder data recognition method the most according to claim 7, it is characterised in that primary election sample With centroid distance more than described average centroid distance μ_pSample, the mark of all samples and average centroid distance in the cluster of calculating Quasi-difference σ_pParticularly as follows:

${\mathrm{\σ}}_p=\sqrt{\frac{1}{N_p}\underset{j=1}{\overset{N_p}{\Σ}}{\left(d\right(j,p)-{\mathrm{\μ}}_p)}^2}---\left(4\right)$

Wherein, μ_pAverage centroid distance in clustering for pth.

A kind of mine gas monitoring disorder data recognition method the most according to claim 1, it is characterised in that described step In S4, for sample belonging to real time data, estimate with normal distribution 95% confidence level, if d is (j, p)-μ_p＞ 1.96 σ_p, the most really It is set to abnormal data sample undetermined.

A kind of mine gas monitoring disorder data recognition method the most according to claim 9, it is characterised in that calculate institute State the difference Δ x of adjacent two elements of exceptional sample undetermined_jIf this vector exists more than historical data sequence maximum difference Δ ax Component, and actual gas monitor value x_jk＞ μ+1.96 σ or x_jk＜ μ-1.96 σ, then be identified as abnormal data.