Mine Ventilation Dynamic Early Warning Method Based on Time-Series Data Analysis

Abstract

The present invention relates to the technical field of data processing, and particularly to a dynamic early warning method for mine ventilation based on time series data analysis. The method obtains the time series of multi-dimensional monitoring data within a historical period; for any multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, the importance degree of each dimension in the multi-dimensional monitoring data is respectively obtained; according to the importance degree of each dimension in the multi-dimensional monitoring data, an optimized COF outlier factor is obtained, and noise data is eliminated according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, so as to obtain the time series of multi-dimensional monitoring data after elimination; a risk prediction model is obtained by using the time series of multi-dimensional monitoring data after elimination, and the dynamic early warning of mine ventilation is carried out by using the risk prediction model. A risk prediction model with higher model accuracy is established through the time series of multi-dimensional monitoring data after noise elimination, and the abnormal early warning identification of mine ventilation that is more in line with the actual situation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a dynamic warning method for mine ventilation based on time series data analysis. Background Art

[0002] The mine ventilation system is an important facility to ensure the safe production of the mine and the health of workers. The parameters that need to be monitored in the mine ventilation system include, but are not limited to, the temperature, humidity, oxygen content, concentration of harmful gases, and wind speed inside the mine. The monitoring of each parameter requires one or more sensors to complete. In addition, the environment of the mine ventilation system changes relatively fast. With the operation of the work and equipment in the mine, parameters such as the temperature, humidity, oxygen content, concentration of harmful gases, and wind speed inside the mine will change, and timely monitoring is required. In order to efficiently monitor the real-time monitoring data in the mine, a neural network model is usually established based on the historical monitoring data of the mine, so as to monitor and warn the abnormal ventilation state in the mine.

[0003] However, because the amount of monitoring data of the mine ventilation system is extremely large, when training the neural network model for warning the abnormal ventilation state in the mine, a large amount of noise data in the monitoring data of the mine ventilation system will cause problems such as a decrease in the accuracy of the model. Therefore, before training the neural network model using the historical monitoring data of the mine ventilation system, a preprocessing process needs to be performed on the historical monitoring data to eliminate the noise data in the historical monitoring data.

[0004] In the prior art, the method for eliminating the noise data in the historical monitoring data is as follows: Obtain the multi-dimensional monitoring data at each historical moment to obtain multi-dimensional monitoring time series data, and evaluate the noise data through the local connectivity differences presented in the multi-dimensional monitoring time series data. The principle is as follows: In the multi-dimensional monitoring time series data, the change of the monitoring data will present a continuous change trend. However, the noise data in the multi-dimensional monitoring time series data will present a relatively high difference from the local change pattern. Therefore, the noise data in the multi-dimensional monitoring time series data is evaluated through the COF outlier factor based on local connectivity.

[0005] However, in the process of evaluating the COF outlier factor based on connectivity for the multi-dimensional monitoring time series data of mine ventilation, for the data points with different timestamps in the multi-dimensional monitoring time series data, the distance between any two data points is measured through the distance difference between them. However, in the process of mine monitoring, the change patterns of different dimensions are different. Therefore, when calculating the distance metric between any two data points, the dimensions with abnormal changes are dragged down by the distance between normal dimensions, resulting in inaccurate evaluation of the noise data in the multi-dimensional monitoring time series data and unable to accurately eliminate the noise data in the monitoring data.

[0006] Therefore, how to improve the accuracy of eliminating noise data in historical monitoring data to enhance the model accuracy of the neural network model for monitoring and warning abnormal ventilation states in a mine has become an urgent problem to be solved. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a mine ventilation dynamic warning method based on time series data analysis to solve the problem of how to improve the accuracy of eliminating noise data in historical monitoring data to enhance the model accuracy of the neural network model for monitoring and warning abnormal ventilation states in a mine.

[0008] Embodiments of the present invention provide a mine ventilation dynamic warning method based on time series data analysis. The method includes the following steps:

[0009] Obtain the preset multi-dimensional monitoring data in the mine environment at each sampling moment, and obtain the multi-dimensional monitoring data time series in the historical period and the monitoring time series data of each dimension.

[0010] Perform STL decomposition on the monitoring time series data of each dimension respectively to obtain the corresponding trend terms. For any multi-dimensional monitoring data in the multi-dimensional monitoring data time series, obtain the trend local windows of each dimension in the multi-dimensional monitoring data respectively from all the trend terms, and obtain the trend change similarity of each dimension in the multi-dimensional monitoring data respectively according to the differences between the trend local windows of each dimension in the multi-dimensional monitoring data.

[0011] Perform normalization processing on the monitoring time series data of each dimension respectively to obtain the corresponding normalized time series data. Obtain the data local windows of each dimension in the multi-dimensional monitoring data respectively from all the normalized time series data, and obtain the importance degree of each dimension in the multi-dimensional monitoring data respectively according to the trend change similarity and the data local windows of each dimension in the multi-dimensional monitoring data.

[0012] Optimize the COF outlier factor of the multi-dimensional monitoring data according to the importance degree of each dimension in the multi-dimensional monitoring data to obtain an optimized COF outlier factor, and eliminate noise data according to the optimized COF outlier factor of each multi-dimensional monitoring data in the multi-dimensional monitoring data time series to obtain an eliminated multi-dimensional monitoring data time series.

[0013] Train a neural network model using the eliminated multi-dimensional monitoring data time series to obtain a risk prediction model, and perform mine ventilation dynamic warning using the risk prediction model.

[0014] Preferably, obtaining the trend change similarity of each dimension in the multi-dimensional monitoring data respectively according to the differences of the trend local windows between each dimension in the multi-dimensional monitoring data includes:

[0015] For any dimension in the multi-dimensional monitoring data, obtain the DTW distance between the trend local window of this dimension and the trend local windows of each dimension other than this dimension in the multi-dimensional monitoring data respectively, normalize the variance of all DTW distances, and the corresponding obtained result is used as the trend change similarity of this dimension.

[0016] Preferably, obtaining the importance degree of each dimension in the multi-dimensional monitoring data respectively according to the trend change similarity and the data local window of each dimension in the multi-dimensional monitoring data includes:

[0017] For any dimension in the multi-dimensional monitoring data, obtain the variance of the trend change similarity according to the trend change similarity of each dimension in the multi-dimensional monitoring data, substitute the opposite number of the variance of the trend change similarity into the exponential function with the natural constant as the base, obtain the corresponding exponential function result, and obtain the first product between the difference between the constant 1 and the exponential function result and the trend change similarity of this dimension;

[0018] Obtain the numerical variance in the data local window of this dimension, perform normalization processing on the numerical variance to obtain the corresponding normalized value, and obtain the second product between the exponential function result, the normalized value and the trend change similarity of this dimension;

[0019] Perform normalization processing on the sum result between the first product and the second product, and the corresponding obtained result is used as the importance degree of this dimension.

[0020] Preferably, optimizing the COF outlier factor of the multi-dimensional monitoring data according to the importance degree of each dimension in the multi-dimensional monitoring data to obtain an optimized COF outlier factor includes:

[0021] Obtain the local window of the multi-dimensional monitoring data in the time series of the multi-dimensional monitoring data, and use the importance degree of each dimension of each data point in the local window of the multi-dimensional monitoring data to obtain the local average linkage distance of the multi-dimensional monitoring data;

[0022] Obtain the optimized COF outlier factor of the multi-dimensional monitoring data according to the local average linkage distance of each data point in the local window of the multi-dimensional monitoring data.

[0023] Preferably, obtaining the local average linkage distance of the multi-dimensional monitoring data by using the importance degree of each dimension of each data point in the local window of the multi-dimensional monitoring data includes:

[0024]

[0025] where d t represents the local average linkage distance of the t-th multi-dimensional monitoring data, K represents the number of data points included in the local window of the t-th multi-dimensional monitoring data, represents the importance degree of the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data, represents the importance degree of the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data, represents the normalized value of the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data, represents the normalized value of the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data, M represents the number of dimensions of the multi-dimensional monitoring data, represents the Euclidean distance between the data corresponding to the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data and the data corresponding to the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data.

[0026] Preferably, eliminating noise data according to the optimized COF outlier factor of each multi-dimensional monitoring data in the multi-dimensional monitoring data time series to obtain the multi-dimensional monitoring data time series after elimination includes:

[0027] For any multi-dimensional monitoring data in the multi-dimensional monitoring data time series, obtain the absolute value of the difference between the optimized COF outlier factor of the multi-dimensional monitoring data and the constant 1, and perform normalization processing on the absolute value of the difference, and the corresponding result is used as the outlier degree of the multi-dimensional monitoring data;

[0028] Obtain a preset outlier degree threshold. If the outlier degree of the multi-dimensional monitoring data is greater than or equal to the outlier degree threshold, determine that the multi-dimensional monitoring data is noise data;

[0029] Obtain all the noise data in the multi-dimensional monitoring data time series, and by performing mean processing on the multi-dimensional monitoring data on both sides of any noise data, replace the corresponding obtained mean with the noise data to obtain the multi-dimensional monitoring data time series after elimination.

[0030] The beneficial effects of the embodiments of the present invention compared with the prior art are:

[0031] The present invention obtains preset multi-dimensional monitoring data in the mine environment at each sampling moment, obtains a time series of multi-dimensional monitoring data within a historical period and monitoring time series data for each dimension; performs STL decomposition on the monitoring time series data for each dimension respectively to obtain corresponding trend terms. For any multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, trend local windows for each dimension in the multi-dimensional monitoring data are respectively obtained from all the trend terms. According to the differences between the trend local windows of each dimension in the multi-dimensional monitoring data, the trend change similarity for each dimension in the multi-dimensional monitoring data is respectively obtained; performs normalization processing on the monitoring time series data for each dimension respectively to obtain corresponding normalized time series data. Data local windows for each dimension in the multi-dimensional monitoring data are respectively obtained from all the normalized time series data. According to the trend change similarity and data local windows for each dimension in the multi-dimensional monitoring data, the importance degree for each dimension in the multi-dimensional monitoring data is respectively obtained; optimizes the COF outlier factor of the multi-dimensional monitoring data according to the importance degree for each dimension in the multi-dimensional monitoring data to obtain an optimized COF outlier factor. Eliminates noise data according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data to obtain an eliminated time series of multi-dimensional monitoring data; trains a neural network model using the eliminated time series of multi-dimensional monitoring data to obtain a risk prediction model, and uses the risk prediction model to perform dynamic warning for mine ventilation. Among them, the importance degree of each dimension is evaluated through the change information of each dimension of each data point in the time series of multi-dimensional monitoring data for mine ventilation, so as to highlight the time series pattern differences between data points during the calculation of the COF outlier factor for each data point in the time series of multi-dimensional monitoring data, accurately identify the noise data in the time series of multi-dimensional monitoring data, and eliminate the noise data. Furthermore, a risk prediction model with higher model accuracy is established through the eliminated time series of multi-dimensional monitoring data for dynamic warning of mine ventilation, realizing a more practical warning identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0033] Figure 1 It is a flowchart of a method for dynamic warning of mine ventilation based on time series data analysis provided in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] Embodiments of the present disclosure will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and intended to explain the present disclosure, and should not be construed as a limitation to the present disclosure.

[0035] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above accompanying drawings are used to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described here can be implemented in an order different from those illustrated or described here. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0036] In order to illustrate the technical solution of the present invention, it will be described below through specific embodiments.

[0037] See Figure 1 , which is a method flowchart of a mine ventilation dynamic warning method based on time series data analysis provided in the first embodiment of the present invention. As Figure 1 shown, the method may include:

[0038] Step S101, obtain the preset multi-dimensional monitoring data in the mine environment at each sampling moment, and obtain the multi-dimensional monitoring data time series in the historical period and the monitoring time series data of each dimension.

[0039] In the mine, the distributed environment monitoring system is used to monitor the environmental conditions of each zone in the mine. For each zone in the mine, a variety of ventilation safety detection instruments are used for environmental monitoring, including but not limited to mine ventilation multi-parameter detectors, dust samplers, respirable dust detectors, oxygen concentration meters, carbon monoxide detectors, air box barometers, etc. A set of environmental monitoring instruments is deployed in each monitoring sub-station in the mine for the distributed environment monitoring system to collect a variety of environmental monitoring data, and the monitoring data is synchronized to the monitoring computer room through the transmission interface in the mine. The monitoring computer room, as the data center in the mine, stores the collected multi-dimensional monitoring data and is used for subsequent analysis.

[0040] In the process of distributed environmental monitoring in a mine, it is necessary to separately analyze data and give early warnings of ventilation status for different monitoring sub-stations (different zones in the mine). In the data center, the monitoring data of different monitoring sub-stations are integrated with time stamps to form a multi-dimensional monitoring data time series of mine ventilation for a single zone. The multi-dimensional monitoring data time series refers to the multi-dimensional monitoring data at multiple sampling moments, and each multi-dimensional monitoring data includes monitoring data of multiple dimensions, and each dimension is a mine ventilation monitoring parameter such as carbon dioxide concentration, temperature, humidity, dust concentration, etc. For the monitoring data of each dimension, the sampling frequency of the corresponding data acquisition device is unified. Thus, based on the preset multi-dimensional monitoring data in the mine environment at each sampling moment, a multi-dimensional monitoring data time series within a historical period can be obtained. Then, the monitoring data of the same dimension in the multi-dimensional monitoring data time series within the historical period is sorted, and the monitoring time series data of each dimension can be correspondingly obtained. Among them, the multi-dimensional monitoring data time series is used as the training data of the BP neural network model for subsequent early warning of abnormal mine ventilation.

[0041] It should be noted that the sampling frequency and the historical period are not restricted, and the implementer can set them according to the implementation scenario.

[0042] Step S102: Perform STL decomposition on the monitoring time series data of each dimension respectively to obtain the corresponding trend terms. For any multi-dimensional monitoring data in the multi-dimensional monitoring data time series, obtain the trend local window of each dimension in the multi-dimensional monitoring data from all the trend terms respectively. According to the differences between the trend local windows of each dimension in the multi-dimensional monitoring data, obtain the trend change similarity of each dimension in the multi-dimensional monitoring data respectively.

[0043] After obtaining the time series of multi-dimensional monitoring data of any monitoring sub-station in the mine, it is necessary to evaluate the noise and optimize the sequence of the obtained multi-dimensional monitoring data time series. Usually, the COF outlier factor can be used for outlier evaluation of time series data. The deviation of the change pattern is evaluated by the local average link distance between the data points in the multi-dimensional monitoring data time series and the data points in the local window, so as to quantify the abnormality of the data points and use it as the outlier evaluation result of the data points. However, for the current mine ventilation scenario, although the outlier evaluation can also be carried out through the measurement of the local average link distance in the time series of multi-dimensional monitoring data for mine ventilation monitoring, in the evaluation of multi-dimensional monitoring data in the mine, there is a problem of ignoring data correlation in the process of calculating the distance between the normalized multi-dimensional monitoring data time series. For example, an increase in temperature may lead to an increase in gas concentration. Correspondingly, in the calculation of the distance between multi-dimensional monitoring data, different weights are required to measure the distance between temperature and gas concentration. However, in the distance measurement process, each dimension with the same weight is calculated, which will cause the dimensions with higher differences to be reduced in difference by similar dimensions, resulting in multi-dimensional monitoring data with a relatively large distance not showing a high outlier evaluation result. Therefore, the traditional connectivity-based local outlier detection algorithm (COF algorithm) has limitations in evaluating the distance between data points in the multi-dimensional monitoring data time series. Therefore, in the embodiments of the present invention, the COF outlier factor of the data points in the multi-dimensional monitoring data time series is optimized to improve the detection of noise data.

[0044] Specifically, first, considering that in the process of evaluating the noise data in the multi-dimensional monitoring data time series, it is necessary to evaluate the trend change similarity and stability of each dimension of each multi-dimensional monitoring data in the multi-dimensional monitoring data time series, so as to more prominently display the dimensions with large differences, which is used to measure the importance of each dimension of each multi-dimensional monitoring data subsequently. Therefore, the STL decomposition is performed on the monitoring time series data of each dimension respectively to obtain the corresponding trend terms. Among them, the STL decomposition belongs to the prior art and will not be elaborated here. Then, for any multi-dimensional monitoring data in the multi-dimensional monitoring data time series, the trend local window of each dimension in the multi-dimensional monitoring data is obtained respectively from all the trend terms. According to the differences between the trend local windows of each dimension in the multi-dimensional monitoring data, the trend change similarity of each dimension in the multi-dimensional monitoring data is obtained respectively. The method for obtaining the trend change similarity of each dimension in the multi-dimensional monitoring data includes:

[0045] For any dimension in the multi-dimensional monitoring data, obtain the DTW distance between the trend local window of this dimension and the trend local window of each dimension other than this dimension in the multi-dimensional monitoring data respectively, normalize the variance of all DTW distances, and the corresponding result is used as the trend change similarity of this dimension.

[0046] In one embodiment, taking the t-th multi-dimensional monitoring data in the time series of multi-dimensional monitoring data as an example, after performing STL decomposition on the monitoring time series data of each dimension to obtain the trend term of each dimension, for the m-th dimension in the t-th multi-dimensional monitoring data, with the t-th data point as the window center in the trend term of the m-th dimension, a trend local window with a window length of 21 is obtained. Similarly, the trend local window of each dimension in the t-th multi-dimensional monitoring data can be obtained. Then, calculate the DTW distance between the trend local window of the m-th dimension in the t-th multi-dimensional monitoring data and the trend local window of each other dimension respectively, so as to obtain the trend similarity of the m-th dimension in the t-th multi-dimensional monitoring data relative to other dimensions as the trend change similarity of the m-th dimension in the t-th multi-dimensional monitoring data. Among them, the DTW distance belongs to the prior art and will not be elaborated here. The calculation expression of the trend change similarity is as follows:

[0047]

[0048] Among them, represents the trend change similarity of the m-th dimension in the t-th multi-dimensional monitoring data, Norm() represents the normalization function, M represents the number of dimensions in the multi-dimensional monitoring data, represents the DTW distance between the trend local window of the m-th dimension in the t-th multi-dimensional monitoring data and the trend local window of the n-th dimension in the t-th multi-dimensional monitoring data, G t,m represents the trend local window of the m-th dimension in the t-th multi-dimensional monitoring data, G t,n represents the trend local window of the n-th dimension in the t-th multi-dimensional monitoring data, represents the mean value of the DTW distances between the trend local window of the m-th dimension in the t-th multi-dimensional monitoring data and the trend local window of each other dimension in the t-th multi-dimensional monitoring data.

[0049] It should be noted that the variance of the DTW distance between the trend local window corresponding to the m-th dimension in the t-th multi-dimensional monitoring data and the trend local windows of other dimensions is used to measure the trend similarity. When the DTW distances between the m-th dimension and other dimensions are uniform, it indicates that in the analysis of importance degree based on similarity, the influence of the m-th dimension is relatively small. However, when the DTW distances between the m-th dimension and other dimensions are non-uniform, it means that the differences between the m-th dimension and some dimensions are large, and some are small, which indicates that the m-th dimension is more important in the local window of the t-th multi-dimensional monitoring data.

[0050] Thus, the trend change similarity of each dimension in the t-th multi-dimensional monitoring data can be obtained.

[0051] Step S103: Normalize the monitoring time series data of each dimension respectively to obtain the corresponding normalized time series data. In all the normalized time series data, obtain the data local windows of each dimension in the multi-dimensional monitoring data respectively. According to the trend change similarity and data local windows of each dimension in the multi-dimensional monitoring data, obtain the importance degree of each dimension in the multi-dimensional monitoring data respectively.

[0052] Since each dimension has different change characteristics at each sampling moment, after obtaining the trend change similarity of each dimension in the t-th multi-dimensional monitoring data, according to the trend change similarity of each dimension in the t-th multi-dimensional monitoring data, evaluate the importance degree of each dimension in the t-th multi-dimensional monitoring data, which is used to accurately measure the link distance between two multi-dimensional monitoring data in the subsequent distance measurement process.

[0053] Considering that when evaluating the importance degree of each dimension in the t-th multi-dimensional monitoring data during the distance measurement process, there is an equilibrium in the differences between multiple dimensions, that is, the trend similarities between each dimension and other dimensions are relatively uniform. At this time, it is necessary to further optimize the importance degree evaluation process through the local stability of each dimension in the t-th multi-dimensional monitoring data to extract more important dimensions. For dimensions with stable local changes, reduce their importance degree, so as to highlight the dimensions with unstable local changes that may be noise. First, use the Norm function to normalize the monitoring time series data of each dimension respectively to obtain the corresponding normalized time series data. Then, in all the normalized time series data, obtain the data local windows of each dimension in the t-th multi-dimensional monitoring data respectively. Specifically, taking the m-th dimension in the t-th multi-dimensional monitoring data as an example, in the normalized time series data of the m-th dimension, with the t-th normalized value as the window center, obtain a data local window with a window length of 21. Similarly, the data local windows of each dimension in the t-th multi-dimensional monitoring data can be obtained.

[0054] After obtaining the data local window for each dimension in the t-th multi-dimensional monitoring data, according to the trend change similarity for each dimension in the t-th multi-dimensional monitoring data and the data local window, the importance degree for each dimension in the t-th multi-dimensional monitoring data is obtained respectively. The method for obtaining the importance degree for each dimension in the t-th multi-dimensional monitoring data is as follows:

[0055] For any dimension in the multi-dimensional monitoring data, according to the trend change similarity for each dimension in the multi-dimensional monitoring data, the variance of the trend change similarity is obtained, and the negative value of the variance of the trend change similarity is substituted into the exponential function with the natural constant as the base to obtain the corresponding exponential function result. The first product between the difference between the constant 1 and the exponential function result and the trend change similarity of the dimension is obtained;

[0056] The numerical variance in the data local window of the dimension is obtained, the numerical variance is normalized to obtain the corresponding normalized value, and the second product between the exponential function result, the normalized value and the trend change similarity of the dimension is obtained;

[0057] The sum result between the first product and the second product is normalized, and the corresponding result is used as the importance degree of the dimension.

[0058] In an embodiment, taking the m-th dimension in the t-th multi-dimensional monitoring data as an example, according to the trend change similarity for each dimension in the t-th multi-dimensional monitoring data, the variance of the trend change similarity is calculated, denoted as Meanwhile, according to all the normalized values included in the m-th dimension, the variance of the normalized values is calculated, and the calculated variance of the normalized values is used as the numerical variance, denoted as σ(g t,m ), and then according to the variance of the trend change similarity and the numerical variance, the importance degree of the m-th dimension in the t-th multi-dimensional monitoring data is obtained. The calculation expression of the importance degree is:

[0059]

[0060] where, ε t,m represents the importance degree of the m-th dimension in the t-th multi-dimensional monitoring data, softmax() represents the normalized exponential function, e represents the natural constant, represents the variance of the trend change similarity for all dimensions in the t-th multi-dimensional monitoring data, Norm() represents the normalization function, σ(g t,m ) represents the variance of the normalized values in the data layout window of the m-th dimension in the t-th multi-dimensional monitoring data, represents the trend change similarity of the m-th dimension in the t-th multi-dimensional monitoring data, and 1 represents the constant.

[0061] It should be noted that The value is used to evaluate the balance of the similarity of trend changes between all dimensions and other dimensions in the t-th multi-dimensional monitoring data. When the distribution is balanced, it means that the weights of all dimensions of the t-th multi-dimensional monitoring data are still equal. Then, further through Norm[σ(g t,m )], the stability evaluation of the corresponding values of the local window of the m-th dimension data of the t-th multi-dimensional monitoring data is carried out. The larger the value of σ(g t,m ), the more unstable the m-th dimension is in the local window of the t-th multi-dimensional monitoring data, the smaller the stability, and the t-th multi-dimensional monitoring data is more likely to be noise data. Correspondingly, it means that the m-th dimension needs to be more prominently presented in the overall distance measurement for noise evaluation of the t-th multi-dimensional monitoring data, that is, the subsequent distance evaluation is carried out with a higher degree of importance; when the distribution is unbalanced, it means that the weights of all dimensions of the t-th multi-dimensional monitoring data are not the same, and the importance degree is evaluated through the similarity of trend changes of the m-th dimension in the t-th multi-dimensional monitoring data. The larger the value, the greater the importance degree of the m-th dimension in the t-th multi-dimensional monitoring data, and the softmax function is used for normalization to ensure that the sum of the importance degrees of all dimensions in the t-th multi-dimensional monitoring data is 1.

[0062] Thus, the importance degree of each dimension in the t-th multi-dimensional monitoring data can be obtained.

[0063] Step S104: Optimize the COF outlier factor of the multi-dimensional monitoring data according to the importance degree of each dimension in the multi-dimensional monitoring data, obtain the optimized COF outlier factor, and eliminate noise data according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of the multi-dimensional monitoring data, so as to obtain the time series of the multi-dimensional monitoring data after elimination.

[0064] Using the above method for obtaining the importance degree of each dimension in the t-th multi-dimensional monitoring data, the importance degree of each dimension of each multi-dimensional monitoring data in the time series of the multi-dimensional monitoring data can be obtained. By utilizing the local connectivity advantage of the COF algorithm and combining with the time series feature differences, the link distance between two multi-dimensional monitoring data is optimized by using the importance degree of each dimension of each multi-dimensional monitoring data in the time series of the multi-dimensional monitoring data, so as to achieve the purpose of accurately evaluating noise data for the multi-dimensional monitoring data of mine ventilation.

[0065] Specifically, according to the importance degree of each dimension in the t-th multi-dimensional monitoring data, the COF outlier factor of the t-th multi-dimensional monitoring data is optimized by optimizing the link distance between two multi-dimensional monitoring data to obtain the corresponding optimized COF outlier factor. The specific optimization process is as follows:

[0066] (1) Obtain the local window of the multi-dimensional monitoring data in the time series of the multi-dimensional monitoring data, and use the importance degree of each dimension of each data point in the local window of the multi-dimensional monitoring data to obtain the local average linkage distance of the multi-dimensional monitoring data.

[0067] Among them, using the importance degree of each dimension of each data point in the local window of the multi-dimensional monitoring data to obtain the local average linkage distance of the multi-dimensional monitoring data includes:

[0068]

[0069] Among them, d t represents the local average linkage distance of the t-th multi-dimensional monitoring data, K represents the number of data points included in the local window of the t-th multi-dimensional monitoring data, represents the importance degree of the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data, represents the importance degree of the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data, represents the normalized value of the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data, represents the normalized value of the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data, M represents the number of dimensions of the multi-dimensional monitoring data, represents the Euclidean distance between the data corresponding to the j-th dimension of the i-th data point in the local window of the t-th multi-dimensional monitoring data and the data corresponding to the j-th dimension of the (i + 1)-th data point in the local window of the t-th multi-dimensional monitoring data.

[0070] It should be noted that in the time series of the multi-dimensional monitoring data, with the t-th multi-dimensional monitoring data as the window center, the 5 data points on each of its left and right sides form the local window of the t-th multi-dimensional monitoring data with a window length of 11.

[0071] (2) According to the local average linkage distance of each data point in the local window of the multi-dimensional monitoring data, obtain the optimized COF outlier factor of the multi-dimensional monitoring data.

[0072] Specifically, after obtaining the local average linkage distance of the t-th multi-dimensional monitoring data, according to the local average linkage distance of each data point in the local window of the t-th multi-dimensional monitoring data, obtain the optimized COF outlier factor of the t-th multi-dimensional monitoring data, and the calculation expression of the optimized COF outlier factor is:

[0073]

[0074] Among them, COF t represents the optimized COF outlier factor of the t-th multi-dimensional monitoring data, K represents the number of data points in the local window of the t-th multi-dimensional monitoring data, and d t represents the local average linkage distance of the t-th multi-dimensional monitoring data, and d o represents the local average linkage distance of the o-th data point in the local window of the t-th multi-dimensional monitoring data.

[0075] It should be noted that the calculation expression of the optimized COF outlier factor is the same as that of the COF outlier factor in the traditional COF algorithm, and will not be elaborated here in detail.

[0076] Similarly, the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data can be obtained. Further, noise data elimination is performed according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, and the time series of multi-dimensional monitoring data after elimination is obtained

[0077] Performing noise data elimination according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data to obtain the time series of multi-dimensional monitoring data after elimination, the specific method is as follows:

[0078] For any multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, obtain the absolute value of the difference between the optimized COF outlier factor of the multi-dimensional monitoring data and the constant 1, and perform normalization processing on the absolute value of the difference. The corresponding result is used as the outlier degree of the multi-dimensional monitoring data;

[0079] Obtain a preset outlier degree threshold. If the outlier degree of the multi-dimensional monitoring data is greater than or equal to the outlier degree threshold, determine that the multi-dimensional monitoring data is noise data;

[0080] Obtain all the noise data in the time series of multi-dimensional monitoring data. By performing mean processing on the multi-dimensional monitoring data on both sides of any noise data, the corresponding obtained mean value is used to replace the noise data, and the time series of multi-dimensional monitoring data after elimination is obtained.

[0081] In an embodiment, the calculation expression of the outlier degree of the t-th multi-dimensional monitoring data is: γ t = Norm(|COF t - 1|), where γ t represents the outlier degree of the t-th multi-dimensional monitoring data, Norm() represents the normalization function, COF t represents the optimized COF outlier factor of the t-th multi-dimensional monitoring data, 1 represents a constant, and || represents the absolute value symbol.

[0082] Using the calculation expression of the outlier degree of the t-th multi-dimensional monitoring data, the outlier degree of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data can be obtained. Then, the outlier degree threshold is set to 0.7. If the outlier degree of any multi-dimensional monitoring data is greater than or equal to 0.7, it is determined that the multi-dimensional monitoring data is noise data. Similarly, all the noise data in the time series of multi-dimensional monitoring data can be obtained.

[0083] For the elimination of noise data, for any noise data, the multi-dimensional monitoring data adjacent to both sides of the noise data is obtained, and the mean value of each dimension is calculated for these two multi-dimensional monitoring data. The values of the same dimension in the noise data are replaced with the mean values of each dimension, so as to obtain the time series of multi-dimensional monitoring data after noise elimination.

[0084] Step S105: Use the time series of multi-dimensional monitoring data after elimination to train the neural network model to obtain a risk prediction model, and use the risk prediction model to conduct dynamic early warning of mine ventilation.

[0085] After obtaining the time series of multi-dimensional monitoring data after noise elimination, the time series of multi-dimensional monitoring data after elimination is used as the training data set of the BP neural network model to train the BP neural network model. The trained BP neural network model is used as the risk prediction model to conduct abnormal early warning of mine ventilation. Among them, the training of the BP neural network model belongs to the prior art and will not be elaborated in detail here. Therefore, after obtaining the risk prediction model, the risk prediction model can be used to conduct dynamic early warning of mine ventilation anomalies. It should be noted that using the risk prediction model to conduct dynamic early warning of mine ventilation is not the focus of this invention and will not be elaborated in detail here.

[0086] In summary, in the embodiment of the present invention, multi-dimensional monitoring data preset in the mine environment at each sampling moment is obtained to obtain a time series of multi-dimensional monitoring data within a historical period and monitoring time series data for each dimension; the monitoring time series data for each dimension is respectively subjected to STL decomposition to obtain corresponding trend terms. For any multi-dimensional monitoring data in the time series of multi-dimensional monitoring data, a trend local window for each dimension in the multi-dimensional monitoring data is respectively obtained from all the trend terms, and according to the differences between the trend local windows of each dimension in the multi-dimensional monitoring data, the trend change similarity for each dimension in the multi-dimensional monitoring data is respectively obtained; the monitoring time series data for each dimension is respectively normalized to obtain corresponding normalized time series data, and a data local window for each dimension in the multi-dimensional monitoring data is respectively obtained from all the normalized time series data. According to the trend change similarity and the data local window for each dimension in the multi-dimensional monitoring data, the importance degree for each dimension in the multi-dimensional monitoring data is respectively obtained; according to the importance degree for each dimension in the multi-dimensional monitoring data, the COF outlier factor of the multi-dimensional monitoring data is optimized to obtain an optimized COF outlier factor, and noise data is eliminated according to the optimized COF outlier factor of each multi-dimensional monitoring data in the time series of multi-dimensional monitoring data to obtain an eliminated time series of multi-dimensional monitoring data; the eliminated time series of multi-dimensional monitoring data is used to train a neural network model to obtain a risk prediction model, and the risk prediction model is used for dynamic early warning of mine ventilation. Among them, the change information of each dimension of each data point in the time series of multi-dimensional monitoring data of mine ventilation is used to evaluate the importance degree of each dimension, so as to highlight the time series pattern differences between data points during the calculation of the COF outlier factor for each data point in the time series of multi-dimensional monitoring data, accurately identify the noise data in the time series of multi-dimensional monitoring data, and eliminate the noise data. Furthermore, a risk prediction model with higher model accuracy is established through the eliminated time series of multi-dimensional monitoring data for dynamic early warning of mine ventilation, realizing a more practical early warning identification.

[0087] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A dynamic early warning method for mine ventilation based on time series data analysis, characterized in that: The method comprises: Obtain the preset multi-dimensional monitoring data in the mine environment at each sampling moment, and obtain the multi-dimensional monitoring data time series sequence and the monitoring time series data of each dimension in the historical period; Perform STL decomposition on the monitoring time series data of each dimension to obtain corresponding trend items. For any multidimensional monitoring data in the multidimensional monitoring data time series sequence, obtain the trend local window of each dimension in the multidimensional monitoring data in all trend items. According to the difference in the trend local windows between the dimensions in the multidimensional monitoring data, obtain the trend change similarity of each dimension in the multidimensional monitoring data. Normalizing the monitoring time series data of each dimension respectively to obtain corresponding normalized time series data, obtaining a data local window of each dimension in the multidimensional monitoring data from all normalized time series data, and obtaining the importance of each dimension in the multidimensional monitoring data according to the trend change similarity of each dimension in the multidimensional monitoring data and the data local window; Optimizing the COF outlier factor of the multidimensional monitoring data according to the importance of each dimension in the multidimensional monitoring data to obtain an optimized COF outlier factor, and performing noise data elimination according to the optimized COF outlier factor of each multidimensional monitoring data in the multidimensional monitoring data time series sequence to obtain a multidimensional monitoring data time series sequence after elimination; The eliminated multi-dimensional monitoring data time series is used to train a neural network model to obtain a risk prediction model, and the risk prediction model is used to perform dynamic early warning of mine ventilation.

2. The mine ventilation dynamic early warning method based on time series data analysis according to claim 1 is characterized in that: The obtaining, based on the difference in trend local windows between the dimensions in the multidimensional monitoring data, the trend change similarity of each dimension in the multidimensional monitoring data includes: For any dimension in the multidimensional monitoring data, the DTW distance between the trend local window of the dimension and the trend local window of each dimension in the multidimensional monitoring data except the dimension is obtained respectively, the variance of all DTW distances is normalized, and the corresponding result is used as the trend change similarity of the dimension.

3. The mine ventilation dynamic early warning method based on time series data analysis according to claim 1 is characterized in that: The obtaining of the importance of each dimension in the multidimensional monitoring data according to the trend change similarity of each dimension in the multidimensional monitoring data and the data local window includes: For any dimension of the multidimensional monitoring data, obtaining a variance of the trend change similarity based on the trend change similarity of each dimension in the multidimensional monitoring data, substituting the inverse of the variance of the trend change similarity into an exponential function with a natural constant as the base to obtain a corresponding exponential function result, and obtaining a first product between a difference between a constant 1 and the exponential function result and the trend change similarity of the dimension; Obtaining a numerical variance in a local window of data of the dimension, normalizing the numerical variance to obtain a corresponding normalized value, and obtaining a second product of the exponential function result, the normalized value, and the trend change similarity of the dimension; A normalization process is performed on the sum of the first product and the second product, and the corresponding result is used as the importance of the dimension.

4. The mine ventilation dynamic early warning method based on time series data analysis according to claim 1 is characterized in that: The step of optimizing the COF outlier factor of the multidimensional monitoring data according to the importance of each dimension in the multidimensional monitoring data to obtain the optimized COF outlier factor comprises: Obtaining a local window of the multidimensional monitoring data in the multidimensional monitoring data time series, and obtaining a local average link distance of the multidimensional monitoring data by using the importance of each dimension of each data point in the local window of the multidimensional monitoring data; An optimized COF outlier factor of the multidimensional monitoring data is obtained according to the local average link distance of each data point in a local window of the multidimensional monitoring data.

5. The mine ventilation dynamic early warning method based on time series data analysis according to claim 4 is characterized in that: The obtaining of the local average link distance of the multidimensional monitoring data by utilizing the importance of each dimension of each data point in the local window of the multidimensional monitoring data comprises: Among them, d t represents the local average link distance of the t-th multidimensional monitoring data, K represents the number of data points contained in the local window of the t-th multidimensional monitoring data, represents the importance of the jth dimension of the ith data point in the local window of the tth multidimensional monitoring data, Indicates the importance of the j-th dimension of the i+1-th data point in the local window of the t-th multidimensional monitoring data, represents the normalized value of the jth dimension of the ith data point in the local window of the tth multidimensional monitoring data, represents the normalized value of the jth dimension of the i+1th data point in the local window of the tth multidimensional monitoring data, M represents the number of dimensions of the multidimensional monitoring data, Represents the Euclidean distance between the data corresponding to the j-th dimension of the i-th data point in the local window of the t-th multidimensional monitoring data and the data corresponding to the j-th dimension of the i+1-th data point in the local window of the t-th multidimensional monitoring data.

6. The mine ventilation dynamic early warning method based on time series data analysis according to claim 1 is characterized in that: The step of performing noise data elimination according to the optimized COF outlier factor of each multidimensional monitoring data in the multidimensional monitoring data time series sequence to obtain the eliminated multidimensional monitoring data time series sequence comprises: For any multidimensional monitoring data in the multidimensional monitoring data time series, obtaining the absolute value of the difference between the optimized COF outlier factor of the multidimensional monitoring data and a constant 1, normalizing the absolute value of the difference, and using the corresponding result as the outlier degree of the multidimensional monitoring data; Obtaining a preset outlier degree threshold, and if the outlier degree of the multidimensional monitoring data is greater than or equal to the outlier degree threshold, determining that the multidimensional monitoring data is noise data; All noise data in the multidimensional monitoring data time series sequence are obtained, and by performing mean processing on the multidimensional monitoring data on both sides of any noise data, the noise data is replaced by the corresponding mean value to obtain the multidimensional monitoring data time series sequence after the noise is eliminated.

Images