Coal mine oil type gas intelligent monitoring method and system based on machine learning

Abstract

This invention discloses a machine learning-based intelligent monitoring method and system for coal mine oil-type gas, comprising the following steps: collecting operating parameters from each monitoring point and arranging them in chronological order to form a multidimensional observation time series; constructing an HDP-HSMM degradation prediction model to establish observation probability distributions and residence time distributions for each implicit degradation state; calculating the probability of entering a dangerous state set and the estimated remaining time based on the observation probability distributions and residence time distributions; comparing the probability of entering a dangerous state set and the estimated remaining time with preset probability thresholds and preset time thresholds, respectively, to generate an early warning signal; generating linkage control commands based on the early warning signals and sending them to ventilation devices, electrical equipment, and personnel evacuation devices; and executing ventilation adjustments, equipment shutdowns, and personnel evacuation operations according to the linkage control commands. This invention achieves intelligent monitoring of coal mine oil-type gas.

Description

Technical Field

[0001] This invention relates to the field of geographic information technology, and in particular to a machine learning-based intelligent monitoring method and system for oil-type gas in coal mines. Background Technology

[0002] In underground coal mines, large quantities of lubricating oil, hydraulic oil, and diesel fuel are used during coal mining, transportation, and the operation of electromechanical equipment. These oil-based gases are easily generated due to high-temperature friction, leakage, and volatilization. When these gases combine with methane, carbon monoxide, and other gases, the risk of fire and explosion increases significantly. Existing coal mine safety monitoring systems primarily rely on traditional gas indicators such as methane and carbon monoxide concentrations, with oil-based gas monitoring often only as a supplement. This often involves a small number of oil-based gas sensors for point-based monitoring, combined with simple upper and lower limit concentration thresholds for alarms. These methods typically focus only on instantaneous concentration values ​​at a single point in time, failing to comprehensively consider the cumulative effects of oil-based gases over time and the impact of changes in operating conditions on their generation and diffusion. Furthermore, they have a low utilization rate of key operating parameters such as equipment load and ventilation changes.

[0003] In terms of data processing and early warning methods, existing technologies generally rely on fixed thresholds and empirical rules for judgment, lacking systematic time-series modeling and multi-state degradation analysis capabilities. On the one hand, oil-type gas concentration is affected by multiple factors such as equipment start-up and shutdown, load fluctuations, and air volume changes, exhibiting obvious randomness and phased variation characteristics. Simple fixed threshold methods are difficult to distinguish between normal fluctuations and early anomalies, easily leading to false alarms and missed alarms. On the other hand, existing methods usually cannot characterize the evolution process of oil-type gas operating conditions from normal and mildly abnormal to severe degradation, and cannot provide the estimated remaining time from the current state to a potentially dangerous state. They can only passively alarm after exceeding the limit, resulting in insufficient foresight and guidance in early warning, making it difficult to provide quantitative decision-making basis for ventilation adjustments, equipment shutdowns, and personnel evacuations.

[0004] Therefore, how to provide intelligent monitoring methods and systems for coal mine oil-type gas based on machine learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a machine learning-based intelligent monitoring method and system for oil-type gas in coal mines. By introducing the HDP-HSMM multi-state degradation prediction model, this invention achieves adaptive modeling and temporal evolution prediction of oil-type gas and related operating conditions. It can provide the probability of entering a dangerous state and the estimated remaining time in advance, reducing the false alarm and false negative rates of traditional threshold methods and improving the accuracy and foresight of early warning. At the same time, it combines early warning signals with automatic responses of ventilation, electrical equipment and personnel evacuation devices, shortening the delay of manual handling and significantly improving the prevention and control of oil-type gas disasters and the level of inherent safety in coal mines.

[0006] The intelligent monitoring method and system for oil-type gas in coal mines based on machine learning according to embodiments of the present invention includes the following steps: Different monitoring points are set up underground in the coal mine. The monitoring points collect oil-type gas concentration, related gas concentration and operating condition parameters synchronized with oil-type gas concentration, and arrange them in chronological order to form a multi-dimensional observation time series. An HDP-HSMM degradation prediction model was constructed, and the number of implicit degradation states of oil-type gas was determined by hierarchical Dirichlet process. The observation probability distribution and residence time distribution of each implicit degradation state were established with operating parameters as conditions, and ordered degradation constraints were applied in the state transition prior. In the real-time monitoring of oil-type gas in coal mines, based on the observation probability distribution and residence time distribution, online inference is performed to obtain the posterior distribution of the implicit degradation state of oil-type gas at the current moment, and the evolution process of the implicit degradation state within the future preset time window is deduced, and the probability of entering the dangerous state set and the estimated remaining time are calculated. The probability of entering a dangerous state set and the estimated remaining time are compared with preset probability thresholds and preset time thresholds, respectively. When any comparison result meets the triggering condition, an early warning signal of the corresponding level is generated. Based on the early warning signal, a linkage control command is generated and sent to ventilation devices, electrical equipment and personnel evacuation devices; The corresponding safety response logic is triggered according to the linkage control command, and ventilation adjustment, equipment shutdown and personnel evacuation operations are performed.

[0007] Optionally, the composition of the multidimensional observation time series specifically includes: Different monitoring points are selected in underground coal mines, including return airways, coal mining faces, areas with concentrated main electrical equipment, conveying roadways, and other areas prone to generating oil-type gas. At each monitoring point, an oil-type gas concentration sensor, related gas concentration sensors, and a working condition parameter acquisition device are installed. The working condition parameters include temperature, wind speed, air volume, equipment load, current, voltage, belt speed, and equipment start-up and shutdown status. Each monitoring point and its sensor are uniformly numbered and configured for communication so that they are connected to the same monitoring system and work according to a unified time reference. A fixed sampling time interval is set for each monitoring point. During the operation of the monitoring system, the oil-type gas concentration sensor collects the oil-type gas concentration value at each sampling time. The relevant gas concentration sensors simultaneously collect the concentration values ​​of carbon monoxide, methane and other preset relevant gases. The operating condition parameters are synchronized by the operating condition parameter acquisition device. The data obtained at the same sampling time at the same monitoring point are combined into a multi-dimensional observation data record and stored according to number and time. For each monitoring point, the multidimensional observation data records are preprocessed, including aligning the data of different monitoring points according to a unified time base, interpolating and filling missing data or marking and removing missing data, filtering out obvious noise and abnormal abrupt values ​​according to preset thresholds and rules, and normalizing or standardizing operating parameters with different dimensions. When global analysis is required, the preprocessed observation data of multiple monitoring points are spliced ​​together according to the monitoring point number order at the same sampling time to generate a global multidimensional observation data record, and arranged in chronological order to form a multidimensional observation time series.

[0008] Optionally, the construction of the HDP-HSMM degradation prediction model specifically includes: For multidimensional observation time series, global lumped hyperparameters and lumped hyperparameters of each layer of the hierarchical Dirichlet process are pre-set. A global weight sequence of implicit degradation states is constructed using a truncation method. By generating the weight values ​​corresponding to each implicit degradation state in sequence and normalizing the weight values, a global state weight vector in an infinite state space is obtained. Implicit degradation states with weight values ​​significantly greater than a preset threshold are identified as oil-type gas implicit degradation states. Each effective implicit degradation state is numbered in sequence from mild degradation to severe degradation to determine the number of implicit degradation states and the initial prior proportion. Based on determining the number and number of hidden degradation states, the operating parameters collected from each monitoring point are used as conditional information. For each numbered oil-type gas hidden degradation state, according to the observation feature description information corresponding to the hidden degradation state, including the oil-type gas concentration, related gas concentration and statistical characteristics of operating parameters, an observation probability distribution parameter set is set for the hidden degradation state. The operating parameters are used as conditional inputs affecting the observation features. A correspondence table from the operating parameters to the expected observation feature interval and trend is established. The correspondence table is associated with the hidden degradation state number, so that when given the operating parameters and the hidden degradation state number, the degree of matching between the current or future observation data and each hidden degradation state can be judged according to the established correspondence. At the same time, a state transition probability prior distribution based on the global state weight vector is set for each numbered hidden degradation state, forming a state transition structure under the hierarchical Dirichlet process framework. For each numbered implicit degradation state of oil-type gas, constraints and descriptions of residence behavior representing the continuous residence time under the current implicit degradation state are defined, including the minimum and maximum allowed continuous residence times. Operating parameters are used as input conditions affecting residence time, and typical residence time ranges for each implicit degradation state under different combinations of operating parameters are specified. At the same time, ordered degradation constraints are applied according to the numerical order of the implicit degradation states, explicitly prohibiting direct transitions from higher-numbered severe degradation states to lower-numbered mild degradation states. Only maintaining within the same implicit degradation state or transitioning stepwise to higher-numbered implicit degradation states is allowed, forming an ordered degradation transition relationship list and residence time constraint rules. Under ordered degradation constraints, the prior distribution of state transition probabilities and the global state weight vector of each implicit degradation state are normalized to form an HDP-HSMM degradation prediction model that satisfies ordered degradation constraints.

[0009] Optionally, the calculation process for the probability of entering the dangerous state set and the estimated remaining time specifically includes: In the real-time monitoring of coal mine oil-type gas, when the data collection at any sampling time is completed, all observation data from the start of monitoring to the current sampling time are extracted from the multidimensional observation time series. The multidimensional observation data corresponding to the current sampling time is marked as the current observation data. The current and historical observation data are input into the HDP-HSMM degradation prediction model in the order of sampling time as input information for online inference. In the HDP-HSMM degradation prediction model, the observed data is recursively calculated time-by-time according to the time sequence. At each sampling time, based on the probability distribution of each hidden degradation state obtained at the previous sampling time and the list of ordered degradation transition relationships between each hidden degradation state, the predicted probability of being in each hidden degradation state at the current sampling time is calculated. At the same time, the observed data at the sampling time is matched with the observation feature description information corresponding to each hidden degradation state, and the matching degree of the current observed data in each hidden degradation state is calculated. The predicted probability is corrected using the matching degree to obtain the updated probability value of being in each hidden degradation state at the sampling time. The normalized value is then normalized and used as the posterior probability distribution of the system in each hidden degradation state at the current sampling time. The posterior probability distribution at the current sampling time is stored as the evaluation result of the current hidden degradation state of oil-type gas. A dangerous state set is formed by selecting representative dangerous latent degenerate states from the set of latent degenerate states. The length of the future prediction time window is predetermined. Taking the posterior probability distribution of the latent degenerate states at the current sampling time as the starting condition, and combining the ordered degeneracy relationship and the corresponding dwell time constraint rules, the probability distribution within the prediction time window is deduced step by step according to the sampling time step. At each future sampling time, the probability value corresponding to each latent degenerate state in the dangerous state set is summarized as the probability of entering the dangerous state set. At the same time, based on the probability distribution of the first entry into the dangerous state set at each future sampling time and the corresponding time interval, the estimated remaining time to enter the dangerous state set is obtained.

[0010] Optionally, the generation of the early warning signal specifically includes: Read the probability of entering the dangerous state set within the preset time window and the estimated remaining time, and call the corresponding preset probability threshold and preset time threshold as a comparison benchmark; The probability of entering the dangerous state set and the estimated remaining time are compared with the preset probability threshold and the preset time threshold, respectively. If the probability of entering the dangerous state set is greater than or equal to the preset probability threshold, or the estimated remaining time is less than or equal to the preset time threshold, the result is marked as meeting the triggering condition. When the triggering conditions are met, the corresponding warning level is determined according to the preset warning classification rules, the probability of entering the dangerous state set, and the expected remaining time. An early warning signal containing the warning level, the probability of entering the dangerous state set, the expected remaining time, and the current monitoring location and time information is generated.

[0011] Optionally, the generation of the linkage control command specifically includes: The system reads the warning level, monitoring location, and time information from the early warning signal. Based on the warning level, it queries the corresponding control strategy in the preset linkage control strategy library to determine the type and basic control parameters of ventilation control, electrical equipment control, and personnel evacuation control to be executed for this warning. It then generates the initial content of the linkage control command. The linkage control strategy library is a set of control rules and control parameters pre-organized according to the warning level, monitoring area, and type of controlled equipment. It is used to store linkage control strategies such as ventilation adjustment, electrical equipment control, and personnel evacuation control corresponding to different warning levels and operating condition combinations. According to the monitoring location, the equipment number, communication address and corresponding control type and control parameters of the ventilation device, electrical equipment and personnel evacuation device that need to be controlled are matched to form a linkage control command containing the target equipment identifier, the type of action to be executed and the action parameters, and then encoded and encapsulated to meet the requirements of the equipment control communication protocol. Through the communication channel of the monitoring system, the linkage control commands are sent to the corresponding ventilation devices, electrical equipment and personnel evacuation devices, and the command sending time, target equipment and execution result information are recorded and archived as linkage control execution records.

[0012] Optionally, the execution of the ventilation adjustment, equipment shutdown, and personnel evacuation operations specifically includes: After the linkage control command is sent to the ventilation device, electrical equipment and personnel evacuation device, each device receives and parses the linkage control command, verifies the device identification, execution action type and action parameters, and when it is confirmed that the command is valid and corresponds to the device, it calls the pre-configured safety response logic and uses the execution action type as the trigger condition for the safety response logic. The ventilation system adjusts the start / stop status, air volume, air speed or air direction of the fan according to the linkage control command, and performs ventilation adjustment operation. The electrical equipment performs shutdown, power cut or load reduction operation according to the linkage control command. The personnel evacuation device activates the functions of sound and light alarm, evacuation instruction or voice broadcast according to the linkage control command, and guides the underground personnel to evacuate to the predetermined safe area. Record the actual action type, start and end time, and execution result of each device. Link the recorded results with the corresponding linkage control command number and warning level for storage, and use them for operation analysis and safety management.

[0013] The intelligent monitoring system for oil-type gas in coal mines based on machine learning according to embodiments of the present invention includes: The underground working condition parameter acquisition module is used to collect oil-type gas concentration, related gas concentration and working condition parameters at different monitoring points in coal mines, and organize them into a multi-dimensional observation time series in chronological order; The HDP-HSMM model building module is used to build an HDP-HSMM degradation prediction model based on multidimensional observation time series, and to establish the observation probability distribution and dwell time distribution of each hidden degradation state. The online degradation state inference module is used to deduce the evolution process of implicit degradation states within a preset time window, and obtain the probability of entering the dangerous state set and the estimated remaining time. The early warning determination and generation module is used to compare the probability of entering a dangerous state set and the estimated remaining time with preset probability thresholds and preset time thresholds, respectively, and generate early warning signals of corresponding levels. The linkage control command generation module is used to select ventilation control, electrical equipment control and personnel evacuation control schemes from the linkage control strategy, generate linkage control commands and send them to the corresponding devices. The safety response execution module is used to perform ventilation adjustment, equipment shutdown or power outage, and personnel evacuation prompts and guidance operations according to the preset safety response logic. The safety response recording module is used to record the process and results of each device performing ventilation adjustments, equipment shutdowns, and personnel evacuation operations, and to store them in association.

[0014] The beneficial effects of this invention are: This invention, through the division of implicit degradation states and ordered degradation constraints, can characterize the gradual evolution process of oil-type gas from normal and slightly abnormal to severely dangerous, thereby achieving refined identification of oil-type gas operating conditions and significantly reducing false alarms and missed alarms caused by simple threshold judgments.

[0015] This invention, by extrapolating the state evolution process within a preset time window, quantifies the probability of entering a set of dangerous states and the estimated remaining time. It can not only determine whether there is a dangerous trend, but also provide the time margin from the current working condition to a potential dangerous state. This overcomes the shortcomings of existing technologies that can only passively alarm after exceeding limits and cannot provide forward-looking time information, and provides quantifiable decision-making basis for ventilation adjustment, equipment shutdown and personnel evacuation.

[0016] This invention directly controls ventilation devices, electrical equipment, and personnel evacuation devices to perform ventilation adjustments, equipment shutdowns, and personnel evacuation operations according to preset safety response logic. It constructs a control system from intelligent monitoring and risk prediction to coordinated response, reducing reliance on human experience and judgment, shortening response time, and significantly improving the automation level and intrinsic safety level of coal mine underground oil and gas disaster prevention and control. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0018] Figure 1 This is a flowchart of the intelligent monitoring method for coal mine oil-type gas based on machine learning proposed in this invention; Figure 2 This is a schematic diagram of the algorithm structure of the intelligent monitoring method for coal mine oil-type gas based on machine learning proposed in this invention. Figure 3 This is a schematic diagram of the module structure of the intelligent monitoring system for coal mine oil-type gas based on machine learning proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1-3 A machine learning-based intelligent monitoring method and system for oil-type gas in coal mines includes the following steps: Different monitoring points are set up underground in the coal mine. The monitoring points collect oil-type gas concentration, related gas concentration and operating condition parameters synchronized with oil-type gas concentration, and arrange them in chronological order to form a multi-dimensional observation time series. An HDP-HSMM degradation prediction model was constructed, and the number of implicit degradation states of oil-type gas was determined by hierarchical Dirichlet process. The observation probability distribution and residence time distribution of each implicit degradation state were established with operating parameters as conditions, and ordered degradation constraints were applied in the state transition prior. In the real-time monitoring of oil-type gas in coal mines, based on the observed probability distribution parameters, online inference is performed to obtain the posterior distribution of the implicit degradation state of oil-type gas at the current moment, and the evolution process of the implicit degradation state within a preset time window is deduced to calculate the probability of entering the dangerous state set. The probability of entering a dangerous state set and the estimated remaining time are compared with preset probability thresholds and preset time thresholds, respectively. When any comparison result meets the triggering condition, an early warning signal of the corresponding level is generated. Based on the early warning signal, a linkage control command is generated and sent to ventilation devices, electrical equipment and personnel evacuation devices; The corresponding safety response logic is triggered according to the linkage control command, and ventilation adjustment, equipment shutdown and personnel evacuation operations are performed.

[0021] In this embodiment, the multidimensional observation time series is specifically composed of: Different monitoring points are selected in underground coal mines, including return airways, coal mining faces, areas with concentrated main electrical equipment, conveying roadways, and other areas prone to generating oil-type gas. At each monitoring point, an oil-type gas concentration sensor, related gas concentration sensors, and a working condition parameter acquisition device are installed. The working condition parameters include temperature, wind speed, air volume, equipment load, current, voltage, belt speed, and equipment start-up and shutdown status. Each monitoring point and its sensor are uniformly numbered and configured for communication so that they are connected to the same monitoring system and work according to a unified time reference. A fixed sampling time interval is set for each monitoring point. During the operation of the monitoring system, the oil-type gas concentration sensor collects the oil-type gas concentration value at each sampling time. The relevant gas concentration sensors simultaneously collect the concentration values ​​of carbon monoxide, methane and other preset relevant gases. The operating condition parameters are synchronized by the operating condition parameter acquisition device. The data obtained at the same sampling time at the same monitoring point are combined into a multi-dimensional observation data record and stored according to number and time. For each monitoring point, the multidimensional observation data records are preprocessed, including aligning the data of different monitoring points according to a unified time base, interpolating and filling missing data or marking and removing missing data, filtering out obvious noise and abnormal abrupt values ​​according to preset thresholds and rules, and normalizing or standardizing operating parameters with different dimensions. When global analysis is required, the preprocessed observation data of multiple monitoring points are spliced ​​together according to the monitoring point number order at the same sampling time to generate a global multidimensional observation data record, and arranged in chronological order to form a multidimensional observation time series.

[0022] In this embodiment, the construction of the HDP-HSMM degradation prediction model specifically includes: For multidimensional observation time series, global lumped hyperparameters and lumped hyperparameters of each layer of the hierarchical Dirichlet process are pre-set. A global weight sequence of implicit degradation states is constructed using a truncation method. By generating the weight values ​​corresponding to each implicit degradation state in sequence and normalizing the weight values, a global state weight vector in an infinite state space is obtained. Implicit degradation states with weight values ​​significantly greater than a preset threshold are identified as oil-type gas implicit degradation states. Each effective implicit degradation state is numbered in sequence from mild degradation to severe degradation to determine the number of implicit degradation states and the initial prior proportion. Based on determining the number and number of hidden degradation states, the operating parameters collected from each monitoring point are used as conditional information. For each numbered oil-type gas hidden degradation state, according to the observation feature description information corresponding to the hidden degradation state, including the oil-type gas concentration, related gas concentration and statistical characteristics of operating parameters, an observation probability distribution parameter set is set for the hidden degradation state. The operating parameters are used as conditional inputs affecting the observation features. A correspondence table from the operating parameters to the expected observation feature interval and trend is established. The correspondence table is associated with the hidden degradation state number, so that when given the operating parameters and the hidden degradation state number, the degree of matching between the current or future observation data and each hidden degradation state can be judged according to the established correspondence. At the same time, a state transition probability prior distribution based on the global state weight vector is set for each numbered hidden degradation state, forming a state transition structure under the hierarchical Dirichlet process framework. For each numbered implicit degradation state of oil-type gas, constraints and descriptions of residence behavior representing the continuous residence time under the current implicit degradation state are defined, including the minimum and maximum allowed continuous residence times. Operating parameters are used as input conditions affecting residence time, and typical residence time ranges for each implicit degradation state under different combinations of operating parameters are specified. At the same time, ordered degradation constraints are applied according to the numerical order of the implicit degradation states, explicitly prohibiting direct transitions from higher-numbered severe degradation states to lower-numbered mild degradation states. Only maintaining within the same implicit degradation state or transitioning stepwise to higher-numbered implicit degradation states is allowed, forming an ordered degradation transition relationship list and residence time constraint rules. Under ordered degradation constraints, the prior distribution of state transition probabilities and the global state weight vector of each implicit degradation state are normalized to form an HDP-HSMM degradation prediction model that satisfies ordered degradation constraints.

[0023] This invention adaptively determines the number and prior proportion of implicit degradation states of oil-type gas within the hierarchical Dirichlet process framework. It introduces an ordered degradation HDP-HSMM degradation prediction model that combines operating parameters with observational features and residence time constraints. This model can automatically mine the statistical characteristics of different degradation stages and reasonably restrict the direction of state transition and continuous residence time, thereby improving the rationality and stability of implicit degradation state classification, enhancing the accuracy of characterizing the multi-state degradation process of oil-type gas, and providing an accurate and reliable model foundation for subsequent online inference and prediction of the probability and remaining time of hazardous states.

[0024] In this embodiment, the calculation process for the probability of entering the dangerous state set and the estimated remaining time specifically includes: In the real-time monitoring of coal mine oil-type gas, when the data collection at any sampling time is completed, all observation data from the start of monitoring to the current sampling time are extracted from the multidimensional observation time series. The multidimensional observation data corresponding to the current sampling time is marked as the current observation data. The current and historical observation data are input into the HDP-HSMM degradation prediction model in the order of sampling time as input information for online inference. In the HDP-HSMM degradation prediction model, the observed data is recursively calculated time-by-time according to the time sequence. At each sampling time, based on the probability distribution of each hidden degradation state obtained at the previous sampling time and the list of ordered degradation transition relationships between each hidden degradation state, the predicted probability of being in each hidden degradation state at the current sampling time is calculated. At the same time, the observed data at the sampling time is matched with the observation feature description information corresponding to each hidden degradation state, and the matching degree of the current observed data in each hidden degradation state is calculated. The predicted probability is corrected using the matching degree to obtain the updated probability value of being in each hidden degradation state at the sampling time. The normalized value is then normalized and used as the posterior probability distribution of the system in each hidden degradation state at the current sampling time. The posterior probability distribution at the current sampling time is stored as the evaluation result of the current hidden degradation state of oil-type gas. A dangerous state set is formed by selecting representative dangerous latent degenerate states from the set of latent degenerate states. The length of the future prediction time window is predetermined. Taking the posterior probability distribution of the latent degenerate states at the current sampling time as the starting condition, and combining the ordered degeneracy relationship and the corresponding dwell time constraint rules, the probability distribution within the prediction time window is deduced step by step according to the sampling time step. At each future sampling time, the probability value corresponding to each latent degenerate state in the dangerous state set is summarized as the probability of entering the dangerous state set. At the same time, based on the probability distribution of the first entry into the dangerous state set at each future sampling time and the corresponding time interval, the estimated remaining time to enter the dangerous state set is obtained.

[0025] This invention obtains the posterior distribution of implicit degradation states by inputting current and historical multidimensional observation data into the HDP-HSMM degradation prediction model. It then combines ordered degradation relationships and residence time constraints to deduce the evolution process of dangerous states within a preset time window, quantitatively calculating the probability of entering a dangerous state set and the estimated remaining time to enter a dangerous state. This enables a dynamic and precise assessment of the risk level and time margin of coal mine oil-type gas, providing a reliable quantitative basis for subsequent early warning and coordinated control.

[0026] In this embodiment, the generation of the early warning signal specifically includes: Read the probability of entering the dangerous state set within the preset time window and the estimated remaining time, and call the corresponding preset probability threshold and preset time threshold as a comparison benchmark; The probability of entering the dangerous state set and the estimated remaining time are compared with the preset probability threshold and the preset time threshold, respectively. If the probability of entering the dangerous state set is greater than or equal to the preset probability threshold, or the estimated remaining time is less than or equal to the preset time threshold, the result is marked as meeting the triggering condition. When the triggering conditions are met, the corresponding warning level is determined according to the preset warning classification rules, the probability of entering the dangerous state set, and the expected remaining time. An early warning signal containing the warning level, the probability of entering the dangerous state set, the expected remaining time, and the current monitoring location and time information is generated.

[0027] In this embodiment, the generation of the linkage control command specifically includes: The system reads the warning level, monitoring location, and time information from the early warning signal. Based on the warning level, it queries the corresponding control strategy in the preset linkage control strategy library to determine the type and basic control parameters of ventilation control, electrical equipment control, and personnel evacuation control to be executed for this warning. It then generates the initial content of the linkage control command. The linkage control strategy library is a set of control rules and control parameters pre-organized according to the warning level, monitoring area, and type of controlled equipment. It is used to store linkage control strategies such as ventilation adjustment, electrical equipment control, and personnel evacuation control corresponding to different warning levels and operating condition combinations. According to the monitoring location, the equipment number, communication address and corresponding control type and control parameters of the ventilation device, electrical equipment and personnel evacuation device that need to be controlled are matched to form a linkage control command containing the target equipment identifier, the type of action to be executed and the action parameters, and then encoded and encapsulated to meet the requirements of the equipment control communication protocol. Through the communication channel of the monitoring system, the linkage control commands are sent to the corresponding ventilation devices, electrical equipment and personnel evacuation devices, and the command sending time, target equipment and execution result information are recorded and archived as linkage control execution records.

[0028] In this embodiment, the execution of ventilation adjustment, equipment shutdown, and personnel evacuation operations specifically includes: After the linkage control command is sent to the ventilation device, electrical equipment and personnel evacuation device, each device receives and parses the linkage control command, verifies the device identification, execution action type and action parameters, and when it is confirmed that the command is valid and corresponds to the device, it calls the pre-configured safety response logic and uses the execution action type as the trigger condition for the safety response logic. The ventilation system adjusts the start / stop status, air volume, air speed or air direction of the fan according to the linkage control command, and performs ventilation adjustment operation. The electrical equipment performs shutdown, power cut or load reduction operation according to the linkage control command. The personnel evacuation device activates the functions of sound and light alarm, evacuation instruction or voice broadcast according to the linkage control command, and guides the underground personnel to evacuate to the predetermined safe area. Record the actual action type, start and end time, and execution result of each device. Link the recorded results with the corresponding linkage control command number and warning level for storage, and use them for operation analysis and safety management.

[0029] A machine learning-based intelligent monitoring system for oil-type gas in coal mines includes: The underground working condition parameter acquisition module is used to collect oil-type gas concentration, related gas concentration and working condition parameters at different monitoring points in coal mines, and organize them into a multi-dimensional observation time series in chronological order; The HDP-HSMM model building module is used to build an HDP-HSMM degradation prediction model based on multidimensional observation time series, and to establish the observation probability distribution and dwell time distribution of each hidden degradation state. The online degradation state inference module is used to deduce the evolution process of implicit degradation states within a preset time window, and obtain the probability of entering the dangerous state set and the estimated remaining time. The early warning determination and generation module is used to compare the probability of entering a dangerous state set and the estimated remaining time with preset probability thresholds and preset time thresholds, respectively, and generate early warning signals of corresponding levels. The linkage control command generation module is used to select ventilation control, electrical equipment control and personnel evacuation control schemes from the linkage control strategy, generate linkage control commands and send them to the corresponding devices. The safety response execution module is used to perform ventilation adjustment, equipment shutdown or power outage, and personnel evacuation prompts and guidance operations according to the preset safety response logic. The safety response recording module is used to record the process and results of each device performing ventilation adjustments, equipment shutdowns, and personnel evacuation operations, and to store them in association.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to an underground area of ​​a coal mine using fully mechanized mining technology. This area contains multiple mining and conveying equipment along the return airway, the coal face, and the main electrical equipment concentration area. During daily operation, large quantities of lubricating oil and diesel fuel are used. During equipment operation, localized heating, oil leakage, and fluctuations in ventilation conditions can easily generate oil-type gases that can combine with methane, carbon monoxide, and other gases to create a complex risk. The mine's original safety monitoring system mainly deployed methane and carbon monoxide sensors, with only a few oil-type gas sensors installed near a few pieces of equipment. This system used a fixed threshold over-limit alarm method, which could not reflect the degradation and evolution of oil-type gases over time, nor could it provide an advance warning of when a dangerous state would occur. There was a lag in the judgment and response of on-site personnel to alarm information.

[0031] In this scenario, multiple monitoring points are arranged along the return airway, working face, and main electrical equipment concentration area. Each monitoring point is equipped with gas sensors for oil-type gas concentration, methane, carbon monoxide, etc., as well as data acquisition devices for operating parameters such as temperature, wind speed, air volume, equipment load, current, voltage, belt speed, and equipment start / stop status. All monitoring points are connected to the same monitoring system, and data is collected using a unified sampling interval and time reference. The monitoring system aligns, interpolates, denoises, and normalizes the data from multiple monitoring points, constructs a multi-dimensional observation time series in chronological order, and inputs this time series into the HDP-HSMM degradation prediction model. Through a hierarchical Dirichlet process, the number and prior proportion of implicit degradation states of oil-type gas are automatically determined, establishing the observation probability distribution and residence time constraints for each implicit degradation state. Ordered degradation constraints are applied in the state transition prior, resulting in a degradation prediction model that meets the requirements of this invention.

[0032] During actual operation, the monitoring system continuously collects oil-type gas and related operating condition data from each monitoring point. It inputs current and historical observation data (within a certain length) into the HDP-HSMM degradation prediction model to calculate the posterior probability distribution of each implicit degradation state at the current moment online. Within a preset time window, it deduces the evolution process of implicit degradation states, obtaining the probability of entering a dangerous state set and the estimated remaining time to enter a dangerous state. The system compares these results with preset probability and time thresholds. When any condition is met, it generates an early warning signal, automatically selects the appropriate ventilation control, electrical equipment control, and personnel evacuation control schemes from the linkage control strategy library, forms linkage control commands, and sends them through the monitoring system to relevant ventilation devices, electrical equipment, and personnel evacuation devices. This triggers safety response logic to execute ventilation adjustments, equipment shutdowns or power outages, and personnel evacuation prompts and guidance.

[0033] Within a continuous monitoring period, statistical analysis was performed on monitoring data under the same monitoring point layout and production conditions. Oil-type gas risk processes verified on-site were selected for comparison. Specific experimental data are shown in Table 1. Table 1. Comparison of traditional fixed threshold methods and the method of this invention.

[0034] As shown in Table 1, under the same 16 monitoring points and 12 confirmed oil-type gas risk processes, the traditional fixed threshold monitoring method resulted in 3 missed reports and 9 false alarms. Furthermore, it failed to provide any early warnings for any of the risk processes and failed to automatically trigger linkage control. In contrast, the HDP-HSMM intelligent monitoring method of the present invention achieved zero missed reports and only 2 false alarms for the 12 risk processes under the same statistical conditions. It has a 100% early warning coverage rate for risk processes, an average early warning time of approximately 22 minutes, and automatically triggered linkage control in 11 of the risk processes. This effectively demonstrates the comprehensive improvement of the present invention in terms of risk identification accuracy, early warning foresight, and linkage response capabilities.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A machine learning-based intelligent monitoring method for oil-type gas in coal mines, characterized in that, Includes the following steps: Different monitoring points are set up underground in the coal mine to collect the operating parameters of each monitoring point and arrange them in chronological order to form a multidimensional observation time series. The operating parameters include temperature, wind speed, air volume, equipment load, current, voltage, belt speed and equipment start-up and shutdown status. An HDP-HSMM degradation prediction model was constructed, and the number of hidden degradation states of oil-type gas was determined by hierarchical Dirichlet process. The observation probability distribution and residence time distribution of each hidden degradation state were established, and ordered degradation constraints were applied in the state transition prior. Based on the observed probability distribution and residence time distribution, online inference is performed to obtain the posterior distribution of the implicit degradation state of oil-type gas at the current moment, and the evolution process of the implicit degradation state within the future preset time window is deduced, and the probability of entering the dangerous state set and the estimated remaining time are calculated. The probability of entering a dangerous state set and the estimated remaining time are compared with preset probability thresholds and preset time thresholds, respectively, to generate an early warning signal; Based on the early warning signal, a linkage control command is generated and sent to ventilation devices, electrical equipment and personnel evacuation devices; The ventilation adjustment, equipment shutdown, and personnel evacuation operations are executed according to the aforementioned linkage control commands. The construction of the HDP-HSMM degradation prediction model specifically includes: For multidimensional observation time series, global lumped hyperparameters and lumped hyperparameters of each layer of the hierarchical Dirichlet process are pre-set. A global weight sequence of implicit degradation states is constructed using a truncation method. By generating the weight values ​​corresponding to each implicit degradation state in sequence and normalizing the weight values, a global state weight vector in an infinite state space is obtained. Implicit degradation states with weight values ​​significantly greater than a preset threshold are identified as oil-type gas implicit degradation states. Each effective implicit degradation state is numbered in sequence from mild degradation to severe degradation to determine the number of implicit degradation states and the initial prior proportion. Based on determining the number and number of latent degradation states, the operating parameters collected from each monitoring point are used as conditional information. For each numbered latent degradation state of oil-type gas, an observation probability distribution parameter set is set according to the observation feature description information corresponding to the latent degradation state. The operating parameters are used as conditional inputs affecting the observation features. A correspondence table from the operating parameters to the expected observation feature interval and change trend is established. At the same time, a state transition probability prior distribution based on the global state weight vector is set for each numbered latent degradation state, forming a state transition structure under the hierarchical Dirichlet process framework. For each numbered implicit degradation state of oil-type gas, constraints and descriptions of residence behavior are defined to represent the continuous residence time under the current implicit degradation state. Operating parameters are used as inputs affecting residence time, and typical residence time ranges for each implicit degradation state under different combinations of operating parameters are specified. At the same time, ordered degradation constraints are applied according to the numerical order of the implicit degradation states, explicitly prohibiting direct transitions from a higher-numbered severe degradation state to a lower-numbered mild degradation state. Only maintaining within the same implicit degradation state or transitioning stepwise to a higher-numbered implicit degradation state is allowed. An ordered degradation transition relationship list and residence time constraint rules are formed. Under ordered degradation constraints, the prior distribution of state transition probabilities and the global state weight vector of each implicit degradation state are normalized to form the HDP-HSMM degradation prediction model.

2. The intelligent monitoring method for coal mine oil-type gas based on machine learning according to claim 1, characterized in that, The multidimensional observation time series specifically includes: Different monitoring points are selected underground in the coal mine. At each monitoring point, an oil-type gas concentration sensor, a related gas concentration sensor, and a working condition parameter acquisition device are installed. Each monitoring point and its sensor are uniformly numbered and configured for communication. A fixed sampling time interval is set for each monitoring point. During the operation of the monitoring system, the oil-type gas concentration sensor collects the oil-type gas concentration value at each sampling time. The relevant gas concentration sensors simultaneously collect the concentration values ​​of carbon monoxide, methane and other preset relevant gases. The operating condition parameters are synchronized by the operating condition parameter acquisition device. The data obtained at the same sampling time at the same monitoring point are combined into a multi-dimensional observation data record and stored according to number and time. For each monitoring point, the multidimensional observation data records are preprocessed and arranged in chronological order to form a multidimensional observation time series.

3. The intelligent monitoring method for coal mine oil-type gas based on machine learning according to claim 1, characterized in that, The calculation process for the probability of entering the dangerous state set and the estimated remaining time specifically includes: In the real-time monitoring of coal mine oil-type gas, when the data collection at any sampling time is completed, all observation data from the start of monitoring to the current sampling time are extracted from the multidimensional observation time series. The multidimensional observation data corresponding to the current sampling time is marked as the current observation data. The current and historical observation data are input into the HDP-HSMM degradation prediction model in the order of sampling time as input information for online inference. In the HDP-HSMM degradation prediction model, the observed data is recursively calculated time-by-time according to the time sequence. At each sampling time, based on the probability distribution of each hidden degradation state obtained at the previous sampling time and the list of ordered degradation transition relationships between each hidden degradation state, the predicted probability of the system being in each hidden degradation state at the current sampling time is calculated. At the same time, the observed data at the sampling time is matched with the observation feature description information corresponding to each hidden degradation state, and the matching degree of the current observed data in each hidden degradation state is calculated. The predicted probability is corrected using the matching degree to obtain the updated probability value, which is then normalized. The normalized result is used as the posterior probability distribution of the system being in each hidden degradation state at the current sampling time. A dangerous state set is formed by selecting representative dangerous latent degenerate states from the set of latent degenerate states. The length of the future prediction time window is predetermined. Taking the posterior probability distribution of the latent degenerate states at the current sampling time as the starting condition, and combining the ordered degeneracy relationship and the corresponding dwell time constraint rules, the probability distribution within the prediction time window is deduced step by step according to the sampling time step. At each future sampling time, the probability value corresponding to each latent degenerate state in the dangerous state set is summarized as the probability of entering the dangerous state set. At the same time, based on the probability distribution of the first entry into the dangerous state set at each future sampling time and the corresponding time interval, the estimated remaining time to enter the dangerous state set is obtained.

4. The intelligent monitoring method for oil-type gas in coal mines based on machine learning according to claim 1, characterized in that, The generation of the early warning signal specifically includes: Read the probability of entering the dangerous state set within the preset time window and the estimated remaining time, and call the corresponding preset probability threshold and preset time threshold as a comparison benchmark; The probability of entering the dangerous state set and the estimated remaining time are compared with the preset probability threshold and the preset time threshold, respectively. If the probability of entering the dangerous state set is greater than or equal to the preset probability threshold, or the estimated remaining time is less than or equal to the preset time threshold, the result is marked as meeting the triggering condition. When the triggering conditions are met, the corresponding warning level is determined according to the preset warning classification rules, the probability of entering the dangerous state set, and the expected remaining time, and an early warning signal is generated.

5. The intelligent monitoring method for oil-type gas in coal mines based on machine learning according to claim 1, characterized in that, The generation of the linkage control command specifically includes: Read the warning level, monitoring location and time information from the early warning signal, query the corresponding control strategy in the preset linkage control strategy library according to the warning level, determine the type and basic control parameters of ventilation control, electrical equipment control and personnel evacuation control to be executed in this warning, and generate the initial content of the linkage control command. According to the monitoring location, the equipment number, communication address, and corresponding control type and control parameters of the ventilation devices, electrical equipment, and personnel evacuation devices that need to be controlled are matched to form linkage control commands, which are then encoded and encapsulated. The linkage control commands are sent to the corresponding ventilation devices, electrical equipment, and personnel evacuation devices through the communication channels of the monitoring system.

6. The intelligent monitoring method for coal mine oil-type gas based on machine learning according to claim 1, characterized in that, The execution of the ventilation adjustment, equipment shutdown, and personnel evacuation operations specifically includes: Each device receives and parses the linkage control command, verifies the device identifier, the type of action to be performed and the action parameters, and calls the pre-configured safety response logic, using the type of action to be performed as the trigger condition for the safety response logic. The ventilation system adjusts the start / stop status, air volume, air speed or air direction of the fan according to the linkage control command, and performs ventilation adjustment operation. The electrical equipment performs shutdown, power cut or load reduction operation according to the linkage control command. The personnel evacuation device activates the functions of sound and light alarm, evacuation instruction or voice broadcast according to the linkage control command, and guides the underground personnel to evacuate to the predetermined safe area. Record the actual action type, start and end time, and execution result of each device. Link the recorded results with the corresponding linkage control command number and warning level for storage, and use them for operation analysis and safety management.

7. A machine learning-based intelligent monitoring system for oil-type gas in coal mines, comprising the machine learning-based intelligent monitoring method for oil-type gas in coal mines as described in any one of claims 1 to 6, characterized in that, include: The underground working condition parameter acquisition module is used to collect oil-type gas concentration, related gas concentration and working condition parameters at different monitoring points in coal mines, and organize them into a multi-dimensional observation time series in chronological order; The HDP-HSMM model building module is used to build an HDP-HSMM degradation prediction model based on multidimensional observation time series, and to establish the observation probability distribution and dwell time distribution of each hidden degradation state. The online degradation state inference module is used to deduce the evolution process of implicit degradation states within a preset time window, and obtain the probability of entering the dangerous state set and the estimated remaining time. The early warning determination and generation module is used to compare the probability of entering a dangerous state set and the estimated remaining time with preset probability thresholds and preset time thresholds, respectively, and generate early warning signals of corresponding levels. The linkage control command generation module is used to select ventilation control, electrical equipment control and personnel evacuation control schemes from the linkage control strategy, generate linkage control commands and send them to the corresponding devices. The safety response execution module is used to perform ventilation adjustment, equipment shutdown or power outage, and personnel evacuation prompts and guidance operations according to the preset safety response logic. The safety response recording module is used to record the process and results of each device performing ventilation adjustments, equipment shutdowns, and personnel evacuation operations, and to store them in association.

Images