Gas concentration inspection and prediction method based on space-time correlation information

Abstract

The application relates to the technical field of gas concentration prediction, in particular to a gas concentration inspection prediction method based on space-time correlation information, which comprises the following steps: collecting methane concentration and wind speed data of each monitoring point; then determining a time sequence dimension emission confidence factor by comparing the time sequence dispersion of single-point data and the change of adjacent periods; meanwhile, calculating a space dimension anti-interference confidence factor by combining the wind speed fluctuation characteristics and the multi-point space distribution; then obtaining a gas emission confidence degree by comprehensively combining the two, and fusing the gas emission confidence degree with physical statistical characteristics of the methane concentration to construct a three-dimensional feature tensor and acquire a prediction model; the application effectively eliminates environmental interference through multi-dimensional confidence correction, and significantly improves the authenticity and accuracy of gas concentration prediction.

Description

Technical Field

[0001] This application relates to the field of gas concentration prediction technology, specifically to a gas concentration inspection and prediction method based on spatiotemporal correlation information. Background Technology

[0002] Methane gas is a common and harmful gas in coal mines, primarily composed of methane. During mining, the breaking of coal seams releases the methane gas originally trapped within the coal, causing it to surge towards the working face. When the methane concentration reaches levels harmful to human health, it not only threatens the health of miners but also poses a significant safety hazard to the mine. Therefore, predicting coal mine methane concentration is of great importance. Existing spatiotemporal methane concentration prediction methods typically combine spatial and temporal models for prediction. For example, spatial attention mechanisms are used to extract spatial features of methane, while temporal attention mechanisms are used to extract temporal features, which are then fused to obtain the spatiotemporal characteristics of methane for concentration prediction.

[0003] However, in actual coal mine scenarios, methane mixes with gases such as oxygen, leading to differences in methane concentrations collected from different monitoring points. This results in varying characterization capabilities of the methane concentration data from each monitoring point, which are also affected by ventilation conditions and change dynamically over time. Existing prediction methods fail to fully consider the differences in the characterization capabilities of gas at different monitoring points and time periods, making it difficult for the extracted spatial and temporal characteristics of gas to fully reflect its true properties, ultimately leading to inaccurate gas concentration prediction results. Summary of the Invention

[0004] In view of the above, it is necessary to provide a gas concentration inspection and prediction method based on spatiotemporal correlation information to solve the above problems.

[0005] One embodiment of this application provides a gas concentration inspection and prediction method based on spatiotemporal correlation information. The method includes: Collect various types of data at each monitoring point at each time within the preset inspection cycle, including methane concentration data and wind speed data; By comparing the methane concentration data and wind speed data at each monitoring point with the data dispersion in each inspection cycle, and combining the changing characteristics of the data dispersion in adjacent inspection cycles, the time-series dimension of the methane concentration data at each monitoring point is determined to have an emergence confidence factor. The fluctuation characteristics of wind speed data at each monitoring point during each inspection cycle are analyzed, and the wind flow oscillation balance is calculated. Based on the same inspection cycle, the overall distribution characteristics of the data dispersion of each type of data at each monitoring point and all monitoring points are analyzed. Combined with the numerical characteristics of the wind flow oscillation balance, the oscillation interference attenuation factor is determined, and the spatial dimension anti-disturbance confidence factor of methane concentration data at each monitoring point in each inspection cycle is obtained. Combining the time-series dimension emission confidence factor and the spatial dimension anti-disturbance confidence factor, the gas emission confidence level is determined. The physical and statistical characteristics of methane concentration data at each monitoring point during each inspection cycle are extracted. Combined with the gas emission confidence level, the inspection cycle input feature vector is obtained. A three-dimensional feature tensor is constructed and a prediction model is obtained for gas concentration prediction.

[0006] The specific process for determining the time-series confidence factor of methane concentration data at each monitoring point is as follows: For each type of data at each monitoring point, the coefficient of variation of each type of data within each inspection cycle is calculated and denoted as the first degree of dispersion and the second degree of dispersion, respectively. Compare the first and second degrees of dispersion obtained for each inspection cycle to determine the significance of outflow within the cycle; Analyze the changes in the first degree of dispersion and the second degree of dispersion obtained in each inspection cycle compared to the previous inspection cycle to obtain the first relative change rate and the second relative change rate respectively; compare the first relative change rate and the second relative change rate to determine the cross-cycle outburst abruptness. When the inspection cycle number is 1, the normalized value of the surge significance within the cycle is used as the surge confidence factor in the time-series dimension; otherwise, the mean of the normalized value of the surge significance within the cycle and the normalized value of the surge mutation degree across cycles is used as the surge confidence factor in the time-series dimension.

[0007] The specific process for obtaining the relative rate of change is as follows: Calculate the difference and maximum value of the methane concentration data between the first degree of dispersion obtained in each inspection cycle and the previous inspection cycle; perform positive fusion on the negative correlation mapping results of the difference and the maximum value to obtain the first relative rate of change; The second relative rate of change is obtained by using the first relative rate of change calculation method, based on wind speed data and the second degree of dispersion.

[0008] Specifically, the calculation of the airflow oscillation balance is as follows: The first-order difference of the sequence of all wind speed data from each monitoring point in each inspection cycle is performed to obtain the wind speed difference sequence. The mean of all positive numbers in the wind speed difference sequence is denoted as the positive wind speed fluctuation intensity; the mean of the absolute values ​​of all negative numbers in the wind speed difference sequence is denoted as the negative wind speed fluctuation intensity. Calculate the sum and absolute value of the difference between the positive and negative wind speed fluctuation intensities; multiply the absolute value of the difference with the negative correlation mapping result of the sum, and then subtract the multiplication result from the value 1 to obtain the wind flow oscillation balance degree.

[0009] The specific process for determining the oscillation interference attenuation factor is as follows: The mean values ​​of all first-degree dispersion and all second-degree dispersion of all monitoring points in each inspection cycle are denoted as the first mean and the second mean, respectively. When the first dispersion of each monitoring point in each inspection cycle is greater than the first mean and the second dispersion is greater than the second mean, if the airflow oscillation balance degree is greater than or equal to a preset value, the difference between value 1 and the airflow oscillation balance degree is calculated, and then multiplied by a preset attenuation intensity parameter. The result is used as the oscillation interference attenuation factor. If the airflow oscillation balance degree is less than the preset value, the difference between value 2 and the preset attenuation intensity parameter is calculated, and then multiplied by the airflow oscillation balance degree. The difference between value 1 and the obtained multiplication result is used as the oscillation interference attenuation factor. Otherwise, the value of 1 will be used as the oscillation interference attenuation factor.

[0010] Specifically, the spatial dimension anti-disturbance confidence factor for obtaining the methane concentration data at each monitoring point in each inspection cycle is as follows: The comparison results of the first dispersion and the first mean of each monitoring point in each inspection cycle are obtained, and then multiplied with the oscillation interference attenuation factor to obtain the spatial dimension anti-disturbance confidence factor.

[0011] Specifically, the gas emission confidence level is the average of the time-series emission confidence factor and the spatial-dimensional disturbance resistance confidence factor.

[0012] The physical statistical features include the mean, maximum value, and linear fitting slope of the sequence of methane concentration data within the inspection cycle; the inspection cycle input feature vector is formed by concatenating the physical statistical features with the corresponding gas emission confidence level.

[0013] Specifically, the construction of the three-dimensional feature tensor is as follows: The inspection cycle input feature vectors of all monitoring points within the current inspection cycle are concatenated in order of monitoring point number to form the global feature of the current inspection cycle; then, all global features within a preset number of consecutive inspection cycles are concatenated in chronological order to form a three-dimensional feature tensor.

[0014] The prediction model is used to output the predicted gas concentration value for the next inspection cycle.

[0015] This application has at least the following beneficial effects: First, this application constructs a time-series-based gas outburst confidence factor by comparing the dispersion of methane concentration and wind speed data at a single monitoring point and their variation characteristics in adjacent periods. This factor can keenly capture the dynamic changes in local gas outburst states, quantifying the reliability of the data from a time perspective. Compared to traditional methods that rely solely on a single threshold for judgment, this application can more accurately identify whether methane concentration fluctuations originate from genuine gas outbursts or are disturbed by drastic wind speed changes, thereby improving the purity of time feature extraction at its source.

[0016] Secondly, this application further analyzes the overall distribution characteristics of wind flow oscillation balance and data dispersion among multiple monitoring points, and constructs a spatial dimension anti-disturbance confidence factor. This factor utilizes spatial correlation to specifically attenuate spurious fluctuations caused by forward and reverse wind flow oscillations. By comprehensively determining the confidence level of gas emission through both temporal and spatial dimensions, the problem of unreasonable weight allocation of data from multiple monitoring points is solved, ensuring that the model can trust real data and suppress noise data in complex wind field environments.

[0017] Finally, instead of simply using confidence level as a filtering threshold, this application combines it with the physicostatistical characteristics of methane concentration to construct a three-dimensional feature tensor input prediction model. This approach preserves the integrity of the physical dimensions of the original data while explicitly injecting data reliability information into the model learning process. This allows the prediction model to not only learn the numerical variation patterns of concentration but also perceive the confidence level of the data, thereby significantly improving the robustness and accuracy of methane concentration prediction. Attached Figure Description

[0018] Figure 1 A flowchart of the gas concentration inspection and prediction method based on spatiotemporal correlation information provided in this application; Figure 2 The gas concentration effect comparison chart provided for this application; Figure 3 A comparison chart of the absolute errors between existing solutions provided for this application and the solution of this application; Figure 4 A comparison chart of the relative errors between existing solutions provided for this application and the solution of this application; Figure 5 A comparison chart of error indices between the existing solutions provided for this application and the solution of this application. Detailed Implementation

[0019] This application proposes a gas concentration inspection and prediction method based on spatiotemporal correlation information, which is applied to the field of gas concentration prediction technology. (See attached document.) Figure 1 The method includes the following steps: Step 1: Collect various types of data at each monitoring point at each time during the preset inspection cycle, including methane concentration data and wind speed data.

[0020] This application collects methane concentration data and wind speed data at different monitoring points and analyzes the importance of gas concentration data at different monitoring points. Specifically, in the coal mine gas inspection system, historical methane concentration data and wind speed data for a preset time period are acquired, where the preset time period is one year in this embodiment; the acquisition frequency is 10Hz, and the duration of one inspection cycle is set to 20s. All methane concentration data in each inspection cycle are arranged in chronological order of acquisition time to obtain a methane concentration sequence; all wind speed data in each inspection cycle are arranged in chronological order of acquisition time to obtain a wind speed sequence.

[0021] During the acquisition of methane concentration and wind speed data, the data may be affected by the surrounding environment, such as people moving around causing local airflow turbulence or vibration of the sensor probe, which may result in noise in the acquired methane concentration and wind speed data. Therefore, this application uses methane concentration and wind speed data as inputs to a Gaussian filtering algorithm for filtering. The standard deviation in the filtering algorithm is set to 1.0, and the filtering window size is set to 5. The Gaussian filtering algorithm is a well-known technology in the field, and the specific calculation method will not be described in detail.

[0022] Step 2: Compare the methane concentration data and wind speed data at each monitoring point to determine the time-series confidence factor of the methane concentration data at each monitoring point in each inspection cycle, and combine the changes in the data dispersion in adjacent inspection cycles.

[0023] Within a single inspection cycle, the degree of fluctuation in methane concentration reflects the extent to which wind speed affects the methane concentration. If wind speed fluctuations are small and methane concentration fluctuations are large, it indicates that the methane concentration fluctuations primarily originate from the actual dynamics of localized gas outbursts. In this case, the temporal characteristic intensity of the methane concentration sequence within that inspection cycle is high. Conversely, if the degree of methane concentration fluctuation changes significantly when wind speed changes gradually between adjacent inspection cycles, it indicates a change in the localized gas outburst state, also corresponding to a high temporal characteristic intensity.

[0024] Ideally, the fluctuations in methane concentration and wind speed at all monitoring points should be consistent within the same inspection cycle. When wind speed fluctuations are constant, if the methane concentration fluctuation is relatively small, it indicates that the monitoring point is strongly affected by wind homogenization, and the methane concentration data has a weak characterizing ability. Conversely, if the methane concentration fluctuation is relatively large, it indicates that the local methane outburst at the monitoring point is abnormally active, and the characterizing ability is strong. It should be noted that both positive and negative wind speed fluctuations will affect the degree of methane concentration fluctuation. For example, when both the relative wind speed fluctuation and the relative methane concentration fluctuation at a monitoring point are large, the deviation in the degree of methane concentration fluctuation should be attenuated according to the scale of the relative wind speed fluctuation.

[0025] Based on the above description, this application will conduct a comparative analysis of methane concentration sequences and wind speed sequences using both single-monitoring-point and multi-monitoring-point data. First, based on the fluctuations and influence relationships between methane concentration data and wind speed data at a single monitoring point, a temporal dimension emission confidence factor is constructed for each monitoring point in each inspection cycle. Then, based on the constraints between methane concentration data and wind speed data at multiple monitoring points, a spatial dimension disturbance resistance confidence factor is constructed for each monitoring point in each inspection cycle. Finally, by combining the temporal dimension emission confidence factor and the spatial dimension disturbance resistance confidence factor, the gas emission confidence level for each monitoring point in each inspection cycle is constructed.

[0026] First, we analyze the importance of methane concentration data for a single monitoring point and a single inspection cycle. We denote the methane concentration sequence and wind speed sequence for the a-th monitoring point and the i-th inspection cycle as L(a,i,1) and L(a,i,2), respectively. Since dispersion can represent the data volatility, we calculate the coefficients of variation of L(a,i,1) and L(a,i,2), and denote them as the first degree of dispersion. Second degree of dispersion When wind speed fluctuations are small and methane concentration fluctuations are large, it indicates that L(a,i,1) has a stronger characterization ability. Therefore, using... The significance of outflow within the period of L(a,i,1) is denoted as ,in To represent extremely small positive numbers, this application aims to prevent the denominator from becoming zero when wind speeds are extremely stable. The value of is subject to special restrictions and can be adjusted according to the actual situation. In this embodiment, The value is set to 0.001. The significance of outbursts within a period reflects the intensity of the fluctuations in L(a,i,1) dominated by localized gas outbursts (i.e., the true intensity of methane fluctuations). The greater the significance of outbursts within a period, the more significant the gas outburst dynamics of L(a,i,1), and the higher the characterization ability and importance of the methane concentration data. It should be further noted that when calculating the coefficient of variation, if the current periodic mean of a sequence is 0, then the coefficient of variation of that sequence is directly assigned a value of 0.

[0027] Furthermore, to analyze the importance of methane concentration data in adjacent inspection cycles at a single monitoring point, in this application, adjacent inspection cycles specifically refer to the current inspection cycle and the previous inspection cycle; the first inspection cycle is not analyzed for adjacent inspection cycles. The methane concentration sequence and wind speed sequence of the (i-1)th inspection cycle at the a-th monitoring point are denoted as L(a,i-1,1) and L(a,i-1,2), respectively. The coefficients of variation of L(a,i-1,1) and L(a,i-1,2) are denoted as... , Since there may be differences in methane concentration and wind speed fluctuations between adjacent inspection cycles, to characterize these differences, the relative change rates of methane and wind speed fluctuations are calculated for the (i-1)th and ith inspection cycles, respectively. Taking the calculation of the first relative change rate of methane fluctuations for the (i-1)th and ith inspection cycles as an example, we know that: max() represents the maximum value function. Let represent the relative rate of change of methane fluctuation level in the i-th and (i-1)-th inspection cycles; correspondingly, let denote the second relative rate of change of wind speed fluctuation level in the i-th and (i-1)-th inspection cycles as . ;when Smaller and A larger value indicates a genuine change in the local gas outburst pattern, even under relatively stable wind speed fluctuations. Therefore, utilizing... The trans-period ejection mutation degree of L(a,i,1) is denoted as ,in, A preset wind speed fluctuation stability threshold is used to prevent numerical overflow caused by an excessively small denominator when the wind flow is extremely stable. In this embodiment... The value is: a lower limit threshold calculated based on the 3σ principle for the relative change of all historical wind speed data; this lower limit threshold is... The cross-cycle outburst variation can reflect the degree of change in methane concentration fluctuation relative to wind speed fluctuation between adjacent inspection cycles. The greater the cross-cycle outburst variation, the higher the degree of change in methane concentration fluctuation under relatively stable wind speed conditions, and the higher the characterization ability of the methane concentration data in the current inspection cycle.

[0028] Similarly, the intra-cycle emission significance and cross-cycle emission abruptness of methane concentration data for all monitoring points in the i-th inspection cycle are calculated. Using the intra-cycle emission significance of methane concentration data for all monitoring points in the i-th inspection cycle as input, the intra-cycle emission significance of methane concentration data for the i-th inspection cycle is normalized based on the maximum value, ensuring that the range of intra-cycle emission significance is within (0,1] to prevent overflow in subsequent calculations. Similarly, the cross-cycle emission abruptness of methane concentration data for all monitoring points in the i-th inspection cycle is normalized.

[0029] Combining the intra-period ejection significance and cross-period ejection abruptness of L(a,i,1), a time-series ejection confidence factor for L(a,i,1) is constructed, with the specific formula as follows: in, The time-series dimension of L(a,i,1) represents the confidence factor. , represents the intra-cycle significance and inter-cycle abrupt change of L(a,i,1), respectively, where i represents the index of the inspection cycle.

[0030] It should be understood that, This can reflect the importance of methane concentration data during the current inspection cycle at a single monitoring point. The larger the value, the more likely the fluctuation in methane concentration in L(a,i,1) originates from a real gas outburst, and the stronger the characterization ability of L(a,i,1). Because and The two methods determine whether the fluctuations in methane concentration in L(a,i,1) under different dimensions belong to the fluctuation state of a real local gas outburst. Therefore, their reliability is the same. and The summation results are then weighted with equal weights.

[0031] To measure the degree to which methane concentration fluctuations are dominated by local outflows from the perspective of a single inspection cycle, The variation of methane concentration fluctuation relative to wind speed fluctuation is measured from the perspective of adjacent inspection cycles. Both characteristics reflect the authenticity of methane concentration fluctuations from different angles. A high value for either characteristic indicates significant gas outbursts in the current or adjacent inspection cycles, effectively improving the reliability of the methane concentration data as originating from actual outburst conditions. Finally, they are fused using an additive approach to more comprehensively evaluate the characterization capability of L(a,i,1). If one characteristic is low, the other characteristic can still play a major role, avoiding excessive weakening of the overall importance due to the inadequacy of a single dimension.

[0032] Step 3: Analyze the fluctuation characteristics of wind speed data at each monitoring point within each inspection cycle, and calculate the wind flow oscillation balance degree; based on the same inspection cycle, analyze the overall distribution characteristics of the data dispersion degree of each type of data at each monitoring point and all monitoring points, and combine the numerical characteristics of the wind flow oscillation balance degree to determine the oscillation interference attenuation factor, and obtain the spatial dimension anti-disturbance confidence factor of methane concentration data at each monitoring point in each inspection cycle; combine the time-series dimension emission confidence factor and the spatial dimension anti-disturbance confidence factor to determine the gas emission confidence degree.

[0033] To analyze the importance of methane concentration data at different monitoring points under the same inspection cycle, firstly, the coefficients of variation for methane concentration and wind speed data at all monitoring points in the i-th inspection cycle are calculated. The first mean of all coefficients of variation obtained from the methane concentration data is denoted as... The second mean of all coefficients of variation obtained from the wind speed data is denoted as . Furthermore, to determine the direction of wind speed fluctuations at all monitoring points, a first-order difference is performed on the wind speed sequence of each monitoring point in the i-th inspection cycle to obtain the wind speed difference sequence for each monitoring point in that inspection cycle. The wind speed difference sequence of the a-th monitoring point in the i-th inspection cycle is denoted as k(a,i,2). The mean of all positive differences in k(a,i,2) is calculated to obtain the positive wind speed fluctuation intensity L(a,i,2), denoted as... If there are no positive differences, the positive wind speed fluctuation intensity is 0; calculate the mean of the absolute values ​​of all negative differences in k(a,i,2) to obtain the negative wind speed fluctuation intensity of L(a,i,2), denoted as . If there is no negative difference, then the negative fluctuation intensity of wind speed is 0; The equilibrium degree of airflow oscillation in L(a,i,2) is denoted as... The wind oscillation balance reflects the equilibrium of wind speed fluctuations in both directions within L(a,i,2). A higher wind oscillation balance indicates that the intensity of the positive and negative fluctuations are closer. In this case, methane concentration fluctuations are more significantly affected by wind speed, and the representation weight of methane concentration data needs to be appropriately reduced. Specifically, when only one-way differences exist in the wind speed difference sequence (i.e.,...) or When the corresponding first-order difference set is empty, the current wind field is determined to be in a monotonically stable state. At this point, the wind oscillation equilibrium degree is directly set... And forcibly set the oscillation interference attenuation factor. .

[0034] If the wind speed remains constant, or only increases / decreases steadily in one direction (without turbulent oscillations), it indicates that the current airflow is very stable. In this environment, if there are drastic fluctuations in methane concentration, it must be caused by a real gas outburst, not a false alarm caused by localized turbulent airflow. In this case, the system should trust the current methane fluctuation data.

[0035] Because different flow field states have different disturbance mechanisms on local gas distribution, when the intensity of forward and reverse wind speed fluctuations is similar, it indicates that the local airflow is in a turbulent state of frequent oscillations, leading to uneven gas mixing and thus causing spurious fluctuations in methane concentration. Conversely, when unidirectional fluctuations dominate or the flow is absolutely stable, it indicates that the flow field is in a stable transition state. In this case, the airflow's interference with gas accumulation is relatively small, and the fluctuations in methane concentration are more likely to reflect the true dynamics of gas emission. Therefore, it is necessary to construct an oscillation disturbance attenuation factor based on the magnitude of the airflow oscillation equilibrium: [The text abruptly ends here, so the translation stops here as well.] , For example, when and At that time, if the airflow oscillates in equilibrium... This indicates that the positive and negative fluctuations in wind speed are similar, resulting in high interference. Therefore, an oscillation interference attenuation factor should be set. ;like This indicates the presence of unidirectional dominance and low interference, thus a damping factor for oscillation interference should be set. In other cases (i.e.) or When, and when the wind speed difference sequence is all zero or only has a one-way difference causing forced (At the same time), it indicates that the airflow has little or no impact on methane concentration, and the oscillation interference attenuation factor can be directly and forcibly set. 。

[0036] in, The attenuation strength parameter represents the attenuation factor of the oscillation interference, preventing... Over-adjustment to the degree of fluctuation in methane concentration, due to At that time, wind speed had a significant impact on the fluctuation of methane concentration, therefore, in this embodiment... The value is 0.6; the oscillation interference attenuation factor reflects the proportion of methane concentration fluctuations that are retained. The smaller the oscillation interference attenuation factor, the stronger the interference of wind speed direction characteristics on methane concentration fluctuations. At this time, the credibility of the real gushing components in the corresponding methane concentration fluctuations is lower. The interference of false fluctuations is suppressed by reducing the weight. Conversely, when the oscillation interference attenuation factor approaches 1, it indicates that the flow field is stable and the original fluctuation characteristics of methane concentration are preserved to the maximum extent.

[0037] Furthermore, the spatial dimension robustness confidence factor of L(a,i,1) is constructed, with the specific formula as follows: in, This represents the spatial dimension robustness confidence factor of L(a,i,1). This represents the oscillation disturbance attenuation factor of L(a,i,1).

[0038] It should be understood that, This reflects the reliability of the relative fluctuation of L(a,i,1) after being disturbed by wind speed during the i-th inspection cycle at all monitoring points. The larger the value, the more significant the relative fluctuations within L(a,i,1), and the more the fluctuations originate from the actual gas outburst dynamics, thus having higher characterization value for the prediction model.

[0039] This ratio can express the degree of difference in methane concentration fluctuation within L(a,i,1) relative to the overall average level of all monitoring points. A ratio greater than 1 indicates that the methane concentration fluctuation at that monitoring point is higher than the overall average level, providing a stronger characterizing effect; a ratio less than 1 indicates that the fluctuation is lower, providing a relatively weaker characterizing effect. However, this ratio may contain spurious fluctuations caused by wind speed oscillations, therefore, a different ratio is introduced. The above ratios are adjusted as weights. Based on the balance of positive and negative fluctuations in wind speed and the dynamic value of whether methane concentration and wind speed are both excessive, the relative fluctuation intensity of methane concentration and the degree of wind speed interference are combined, which can more accurately reflect the true characterization ability of L(a,i,1) for gas outburst dynamics.

[0040] After completing all historical inspection cycles , After feature calculation, their global maximum values ​​in all historical inspection cycles are extracted and globally normalized. Based on the normalized data, the confidence scores of all gas outbursts are calculated, and the global maximum values ​​of the gas outburst confidence scores are extracted and globally normalized. These three global maximum value parameters are saved for real-time prediction normalization in the new inspection cycle.

[0041] Furthermore, by combining the temporal dimension emission confidence factor and the spatial dimension disturbance resistance confidence factor of L(a,i,1), the gas emission confidence score of L(a,i,1) is constructed, and the specific formula is as follows: Represents the confidence level of gas outburst for L(a,i,1); due to and The two methods, which calculate the fluctuation of methane concentration under different dimensions without the interference of wind speed, describe the characterization ability of monitoring points within the same inspection cycle from different perspectives. Therefore, this application uses the same weight for weighted processing.

[0042] It should be understood that, This comprehensively reflects L(a,i,1)'s ability to accurately represent the dynamics of gas outburst. The larger the value, the less affected L(a,i,1) is by wind speed, and the greater the fluctuation of the true methane concentration it represents. Therefore, it should be given a higher weight in subsequent data fusion.

[0043] Similarly, the gas emission confidence scores of historical methane concentration data for all monitoring points in all inspection cycles are calculated. Using the gas emission confidence scores of historical methane concentration data for all monitoring points in all inspection cycles as input, the gas emission confidence scores of historical methane concentration data for all monitoring points in all inspection cycles are normalized based on the maximum value normalization method, so that the range of gas emission confidence scores is standardized between (0,1], which facilitates subsequent data fusion and prevents data overflow.

[0044] Step 4: Extract the physical and statistical characteristics of methane concentration data at each monitoring point within each inspection cycle, combine them with the gas emission confidence level to obtain the inspection cycle input feature vector, construct a three-dimensional feature tensor and obtain a prediction model for gas concentration prediction.

[0045] For each inspection cycle in the historical data, the original data is no longer weighted or altered. Instead, the mean, maximum, and linear fitting slope of the original methane concentration data sequence for each monitoring point are directly extracted as three basic physical statistics reflecting the gas concentration in that inspection cycle. These three physical statistics are concatenated with the gas emission confidence score calculated for that monitoring point in that inspection cycle to form a 4-dimensional inspection cycle input feature vector for that monitoring point. The inspection cycle input feature vectors of all monitoring points within the current inspection cycle are concatenated in order of monitoring point number to form the global features for that inspection cycle. Then, all global features within a continuous time window (e.g., a time window length of 10 inspection cycles) are concatenated in chronological order to form a three-dimensional tensor (number of samples × time window length × global feature dimension per inspection cycle), which is used as input to a two-layer Long Short-Term Memory (LSTM) network for model training. The network has 64 hidden units in the first layer, 32 hidden units in the second layer, and a fully connected output layer to predict the average gas concentration in the next inspection cycle (20 seconds). The loss function is mean squared error loss, the optimizer is Adam, the initial learning rate is 0.001, the batch size is 32, and the number of training rounds is 100.

[0046] For real-time collected data from new inspection cycles, the mean, maximum, and linear fitting slope of the original methane concentration data sequence for each monitoring point in the current inspection cycle are first extracted. Simultaneously, the gas emission confidence level for each monitoring point in the current inspection cycle is calculated, and the global maximum value saved during the historical training phase is used to normalize the confidence level for the current inspection cycle in real time. The three extracted physical statistics are concatenated with the real-time normalized confidence levels to obtain the current inspection cycle input feature vector for each monitoring point. This vector is then concatenated in the same way as during the training phase and input into the trained LSTM prediction model to output the predicted gas concentration value for the next inspection cycle.

[0047] To verify the effectiveness of the prediction model in this application, a comparative experiment was conducted with an existing scheme that does not incorporate fusion weights. Figure 2 The comparison chart of gas concentration effects shows that the predicted values ​​of this application's scheme have a higher degree of fit with the actual gas concentration curve, and can accurately track local concentration mutations. Combined with... Figure 3 The absolute error comparison chart and Figure 4 As can be seen from the relative error comparison chart, the prediction error range of the proposed solution is significantly reduced, and the error fluctuation is more stable. Combined with... Figure 5 As shown in the error index comparison chart, the proposed solution outperforms existing solutions in all conventional quantitative error assessment indicators (such as root mean square error and mean absolute error). In summary, these results demonstrate that by introducing a fusion weight based on the relationship between methane and wind speed fluctuations, the proposed solution effectively suppresses spurious fluctuation interference caused by wind speed oscillations and improves the accuracy of methane concentration prediction.

Claims

1. A gas concentration inspection and prediction method based on spatiotemporal correlation information, characterized in that, The method includes the following steps: Collect various types of data at each monitoring point at each time within the preset inspection cycle, including methane concentration data and wind speed data; By comparing the methane concentration data and wind speed data at each monitoring point with the data dispersion in each inspection cycle, and combining the changing characteristics of the data dispersion in adjacent inspection cycles, the time-series dimension of the methane concentration data at each monitoring point is determined to have an emergence confidence factor. The fluctuation characteristics of wind speed data at each monitoring point during each inspection cycle are analyzed, and the wind flow oscillation balance is calculated. Based on the same inspection cycle, the overall distribution characteristics of the data dispersion of each type of data at each monitoring point and all monitoring points are analyzed. Combined with the numerical characteristics of the wind flow oscillation balance, the oscillation interference attenuation factor is determined, and the spatial dimension anti-disturbance confidence factor of methane concentration data at each monitoring point in each inspection cycle is obtained. Combining the time-series dimension emission confidence factor and the spatial dimension anti-disturbance confidence factor, the gas emission confidence level is determined. The physical and statistical characteristics of methane concentration data at each monitoring point during each inspection cycle are extracted. Combined with the gas emission confidence level, the inspection cycle input feature vector is obtained. A three-dimensional feature tensor is constructed and a prediction model is obtained for gas concentration prediction.

2. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The specific process for determining the time-series confidence factor of methane concentration data at each monitoring point is as follows: For each type of data at each monitoring point, the coefficient of variation of each type of data within each inspection cycle is calculated and denoted as the first degree of dispersion and the second degree of dispersion, respectively. Compare the first and second degrees of dispersion obtained for each inspection cycle to determine the significance of outflow within the cycle; Analyze the changes in the first degree of dispersion and the second degree of dispersion obtained in each inspection cycle compared to the previous inspection cycle to obtain the first relative change rate and the second relative change rate respectively; compare the first relative change rate and the second relative change rate to determine the cross-cycle outburst abruptness. When the inspection cycle number is 1, the normalized value of the surge significance within the cycle is used as the surge confidence factor in the time-series dimension; otherwise, the mean of the normalized value of the surge significance within the cycle and the normalized value of the surge mutation degree across cycles is used as the surge confidence factor in the time-series dimension.

3. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 2, characterized in that, The specific process for obtaining the relative rate of change is as follows: Calculate the difference and maximum value of the methane concentration data between the first degree of dispersion obtained in each inspection cycle and the previous inspection cycle; perform positive fusion on the negative correlation mapping results of the difference and the maximum value to obtain the first relative rate of change; The second relative rate of change is obtained by using the first relative rate of change calculation method, based on wind speed data and the second degree of dispersion.

4. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The calculation of the airflow oscillation balance is specifically as follows: The first-order difference of the sequence of all wind speed data from each monitoring point in each inspection cycle is performed to obtain the wind speed difference sequence. The mean of all positive numbers in the wind speed difference sequence is denoted as the positive wind speed fluctuation intensity. The mean of the absolute values ​​of all negative numbers in the wind speed difference sequence is denoted as the intensity of the negative wind speed fluctuation. Calculate the sum and absolute value of the difference between the positive and negative wind speed fluctuation intensities; multiply the absolute value of the difference with the negative correlation mapping result of the sum, and then subtract the multiplication result from the value 1 to obtain the wind flow oscillation balance degree.

5. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 2, characterized in that, The specific process for determining the oscillation interference attenuation factor is as follows: The mean values ​​of all first-degree dispersion and all second-degree dispersion of all monitoring points in each inspection cycle are denoted as the first mean and the second mean, respectively. When the first dispersion of each monitoring point in each inspection cycle is greater than the first mean and the second dispersion is greater than the second mean, if the airflow oscillation balance is greater than or equal to the preset value, calculate the difference between the value 1 and the airflow oscillation balance, and then multiply it by the preset attenuation intensity parameter. The result is used as the oscillation interference attenuation factor. If the airflow oscillation balance is less than the preset value, calculate the difference between the value 2 and the preset attenuation intensity parameter, and then multiply it by the airflow oscillation balance; use the difference between the value 1 and the obtained multiplication result as the oscillation interference attenuation factor. Otherwise, the value of 1 will be used as the oscillation interference attenuation factor.

6. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 5, characterized in that, The spatial dimension anti-disturbance confidence factor for obtaining the methane concentration data at each monitoring point in each inspection cycle is specifically as follows: The comparison results of the first dispersion and the first mean of each monitoring point in each inspection cycle are obtained, and then multiplied with the oscillation interference attenuation factor to obtain the spatial dimension anti-disturbance confidence factor.

7. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The gas emission confidence level is specifically the average of the time-series emission confidence factor and the spatial-dimensional disturbance resistance confidence factor.

8. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The physical statistical features include: the mean, maximum value, and linear fitting slope of the sequence of methane concentration data within the inspection cycle; the input feature vector of the inspection cycle is formed by concatenating the physical statistical features with the corresponding gas emission confidence.

9. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The construction of the three-dimensional feature tensor is specifically as follows: The inspection cycle input feature vectors of all monitoring points within the current inspection cycle are concatenated in order of monitoring point number to form the global feature of the current inspection cycle; then, all global features within a preset number of consecutive inspection cycles are concatenated in chronological order to form a three-dimensional feature tensor.

10. The gas concentration inspection and prediction method based on spatiotemporal correlation information as described in claim 1, characterized in that, The prediction model is used to output the predicted gas concentration value for the next inspection cycle.

Images