Method and system for automatically finding and verifying coal and gas outburst early warning criteria

Abstract

The present application discloses a method and system for automatically finding and verifying coal and gas outburst warning criteria, which relates to the field of coal and gas outburst criterion finding, including: obtaining physical information and engineering operation information of the target area, and preprocessing to generate target features; based on the target features, using a machine learning model to perform risk assessment and generate risk assessment information; based on the target features, initializing the first iterative model; adjusting the model parameters of the first iterative model based on the risk assessment information, and iteratively finding the target warning criterion; in response to the first iterative model reaching the maximum number of iterations, or the fitness of the warning criterion during the iteration process being greater than the target threshold, stopping the iteration and outputting the current warning criterion as the target warning criterion. In this way, the applicability of the warning criterion can be improved and the engineering activities can be better combined with the physical information of the coal mine area, providing reliable criteria for subsequent outburst prediction and monitoring.

Description

Technical Field

[0001] The present application relates to the field of coal and gas outburst criterion search, and in particular to a method and system for automatically searching and verifying coal and gas outburst early warning criterion. Background Art

[0002] With the development of sensing technology, data processing technology and artificial intelligence technology, multi-sensor fusion and data-driven coal mine safety monitoring methods have gradually received attention. These methods integrate real-time data collected by multiple sensors and use data analysis algorithms and machine learning models to make more accurate and real-time assessments of coal and gas outburst warnings. However, the existing multi-sensor fusion warning system still has some shortcomings. For example, it relies on a single or a small number of indicators, such as gas concentration and gas outburst volume, and cannot fully capture the complex changes in the coal mine environment; lacks an automated warning criterion generation and verification mechanism, making it difficult to achieve real-time and dynamic warning adjustments, and cannot respond to environmental changes in a timely manner; the relationship between engineering operations and coal and gas outbursts has not been deeply explored, and can only be responded to passively.

[0003] For example, the Chinese patent with the authorization announcement number CN117489413B discloses a method for advanced detection and early warning of coal seam gas outburst anomalies, establishes a correlation model between coal seam gas multi-parameters and gas outburst volume, and measures coal seam gas parameters such as coal seam residual gas content, coal seam residual gas pressure, coal body solidity coefficient, and drill cuttings gas desorption index at multiple points during the excavation process while constructing ultra-long directional drilling holes along the coal seam. The parameter measurement results and underground tunnel air volume and gas concentration sensor data are uploaded to the ground big data center in real time. The ground big data center makes intelligent analysis and prediction of abnormal gas outbursts in front of the working face based on the gas outburst volume model, historical database, and machine learning algorithm, and promptly issues an alarm to the underground, achieving the purpose of advanced prediction and early warning of coal seam gas outburst anomalies. This invention can make advance predictions and early warnings of future gas outburst anomalies in front of the working face, ensuring rapid and safe excavation of coal mine tunnels.

[0004] The above methods have the problems raised by this background technology, namely, they rely on a small number of indicators, such as gas concentration, coal body solidity coefficient, drill cuttings gas desorption index, etc., and are unable to capture the complex changes in the coal mine environment in real time and comprehensively. They do not deeply explore the relationship between engineering operations and coal and gas outbursts, and focus on using gas parameters as the main factor in evaluating the danger of outbursts, which violates the recognized coal and gas outburst mechanism - outbursts are "stimulated" by ground stress and "driven" by gas; ground stress controls gas, and without the stimulation of coal seams by ground stress (strong impact on coal seams), there will be no gas driving phenomenon.

[0005] The research results show that under the action of original stress, most of the gas in the coal seam is stored in the microscopic void structure of the coal seam in an adsorbed state. This adsorbed gas has no energy to drive the outburst. Only when the top rock layer under the ground stress suddenly breaks, a strong impact force is generated on the coal seam. Under the impact force, the original coal seam is instantly broken into powder, and the permeability of the coal seam suddenly changes, causing the adsorbed gas to desorb - from the adsorbed state to the free state, expanding in the coal seam, and then generating the energy to drive the outburst. Using gas parameters as the main analysis index only explores the appearance of the outburst, not the essence. This technology cannot meet the needs of the site in time and space for timely discovery of the precursor information of the outburst danger and early warning. Moreover, the drill cuttings gas desorption index has almost no real-time outburst danger warning due to the discontinuous test (tested every few days), and cannot solve the current problem of real-time tracking and warning of outburst danger in domestic outburst mines. Summary of the invention

[0006] In view of the shortcomings of the existing technology, this application discloses a method and system for automatically finding and verifying coal and gas outburst early warning criteria, which can effectively solve the problems in the existing technology that the early warning criteria are relatively single, cannot be dynamically adjusted, and are not deeply integrated with engineering operations. The specific scheme is as follows:

[0007] In the first aspect, the present application provides a method for automatically finding and verifying coal and gas outburst early warning criteria, including:

[0008] Obtain physical information and engineering operation information of the target area, perform preprocessing, and generate target features;

[0009] Based on the target characteristics, a machine learning model is used to perform risk assessment to generate risk assessment information; wherein the machine learning model is constructed based on historical physical information and historical engineering operation information of the target area;

[0010] Based on the target features, a first iterative model is initialized; based on the risk assessment information, model parameters of the first iterative model are adjusted, and a target early warning criterion is iteratively searched; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process is greater than a target threshold, the iteration is stopped and the current early warning criterion is output as the target early warning criterion.

[0011] As an optional implementation manner, the target feature includes: a first feature, a second feature, and a third feature; and generating the target feature includes:

[0012] For the physical information, feature extraction is performed based on the wavelet transform method to generate a first feature for characterizing the mutation characteristic; feature extraction is performed based on the Fourier transform method to generate a second feature for characterizing the periodic characteristic;

[0013] Based on the physical information and the engineering operation information, a third feature is generated to characterize the dependency between the physical characteristics of the target area and the engineering activities.

[0014] As an optional implementation manner, the generating, based on the physical information and the engineering operation information, a third feature for characterizing the dependency relationship between the physical characteristics of the target area and the engineering activity comprises:

[0015] Performing time stamp alignment on the physical information and engineering operation information of the target area;

[0016] Based on the engineering operation information, determine target operation information, and divide a time window for each target operation information;

[0017] Determining physical characteristic change information based on the physical information of the target area within each of the time windows;

[0018] The third feature is generated based on the physical feature change information within each of the time windows and the target operation information corresponding to each of the time windows.

[0019] As an optional implementation manner, the third feature includes: an intersection feature and a risk feature; the generating the third feature based on the physical feature change information within each of the time windows and the target operation information corresponding to each of the time windows includes:

[0020] In each of the time windows, based on the physical feature change information and the target job information, create the cross-feature between the physical feature statistics and the target job;

[0021] In each of the time windows, based on the physical feature change information and the target operation information, a statistical threshold method is used to determine potential risk information and generate the risk feature.

[0022] As an optional implementation, initializing the first iterative model based on the target feature includes:

[0023] Based on the target features, an initial population is generated, wherein each individual represents an early warning criterion, wherein the early warning criterion includes at least one of the target features.

[0024] As an optional implementation, each of the individuals further includes at least one characteristic parameter of the target characteristic included in the individual.

[0025] As an optional implementation, the first iterative model includes: a whale optimization sub-model and a non-dominated sorting genetic sub-model; the iterative search for target warning criterion combination includes:

[0026] Based on the physical information and engineering operation information of the target area, construct a fitness function, wherein the fitness function is used to evaluate at least one of the following performance indicators: warning accuracy, false alarm rate, response speed, and computing cost;

[0027] In each iteration cycle of the first iteration model, the whale optimization sub-model is used to update the position of each of the individuals, wherein the position update of each of the individuals is used to update the characteristic parameters that characterize the target characteristics;

[0028] The non-dominated sorting genetic sub-model is used to perform non-dominated sorting on the individuals updated by the whale optimization sub-model, determine the dominance level of each individual, and calculate the crowding distance of each individual; based on the dominance level of each individual and the crowding distance of each individual, select the parent individual and perform crossover and mutation operations to generate new offspring individuals.

[0029] As an optional implementation manner, in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stopping iteration and outputting the current early warning criterion as a target early warning criterion includes:

[0030] In response to the generation of a new offspring individual, the fitness of the new offspring individual is calculated based on the fitness function;

[0031] In response to the fitness of the new offspring individual being greater than the target threshold, the iteration is stopped and the current warning criterion is output as the target warning criterion; otherwise, the new offspring individual is combined with the initial population as the population of the first iterative model in the next round of iteration.

[0032] As an optional implementation manner, adjusting the model parameters of the first iterative model based on the risk assessment information includes:

[0033] Based on the risk assessment information, adjusting population information, search intensity parameters, spiral update parameters, and random search probability in the whale optimization sub-model;

[0034] Based on the risk assessment information, the population information, crossover rate, and mutation rate in the non-dominated sorting genetic submodel are adjusted.

[0035] In the second aspect, the present application provides a system for automatically finding and verifying coal and gas outburst early warning criteria, including:

[0036] The acquisition unit is used to obtain the physical information and engineering operation information of the target area, and perform preprocessing to generate target features;

[0037] An assessment unit, configured to perform risk assessment based on the target characteristics using a machine learning model to generate risk assessment information; wherein the machine learning model is constructed based on historical physical information and historical engineering operation information of the target area;

[0038] An iteration unit is used to initialize a first iterative model based on the target feature; adjust the model parameters of the first iterative model based on the risk assessment information, and iteratively search for a target early warning criterion; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stop iteration and output the current early warning criterion as the target early warning criterion.

[0039] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory, and when the machine-readable instructions are executed by the processor, the machine-readable instructions perform the steps of the above-mentioned first aspect, or any possible implementation of the first aspect.

[0040] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are executed.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] By collecting a variety of physical information and engineering operation information in the target area, and performing multiple feature extractions such as wavelet transform and Fourier transform, the complex changes in the coal mine environment and the impact of human engineering activities on the coal mine environment can be fully captured;

[0043] Based on real-time risk assessment information, model parameters are adjusted dynamically to respond to environmental changes in real time, thereby improving the response speed and adaptability of the early warning system;

[0044] By combining the whale optimization algorithm (WOA) and the non-dominated sorting genetic algorithm (NSGA-II), the present invention automatically searches for and verifies the best combination of early warning criteria during the iterative optimization process, realizing the organic combination of global search and local optimization. The global search capability of WOA can quickly explore a wide range of solution spaces and discover potential excellent solutions, while the multi-objective optimization characteristics of NSGA-II effectively balance the false alarm rate while ensuring the accuracy of early warnings through non-dominated sorting and crowding distance calculations. This combination not only enhances the robustness and adaptability of the algorithm, enabling it to better cope with the complex and changeable coal mine environment, but also improves the optimization efficiency, accelerates the convergence speed, and significantly improves the accuracy and reliability of early warning criteria. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flow chart of the method for automatically finding and verifying coal and gas outburst early warning criteria provided for this application;

[0046] Figure 2 A schematic diagram of a method for automatically finding and verifying coal and gas outburst early warning criteria provided in this application;

[0047] Figure 3 A schematic diagram of the structure of the automatic search and verification system for coal and gas outburst early warning criteria provided for this application;

[0048] Figure 4 This is a structural diagram of the electronic device provided for this application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0050] See also Figure 1 and Figure 2 , Figure 1 This is a flow chart of the method for automatically finding and verifying coal and gas outburst early warning criteria of the present invention. Figure 2 A schematic diagram of a method for automatically finding and verifying coal and gas outburst early warning criteria provided in this application, the method includes:

[0051] S101, obtaining physical information and engineering operation information of the target area, and performing preprocessing to generate target features;

[0052] S102, based on the target characteristics, using a machine learning model to perform risk assessment and generate risk assessment information; wherein the machine learning model is constructed based on historical physical information and historical engineering operation information of the target area;

[0053] S103. Initialize a first iterative model based on the target features; adjust the model parameters of the first iterative model based on the risk assessment information, and iteratively search for target warning criteria; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the warning criterion during the iteration process being greater than a target threshold, stop iteration and output the current warning criterion as the target warning criterion.

[0054] Regarding S101 above:

[0055] In practice, the target area refers to a specific working face or area in the mine that needs to be monitored and analyzed. In coal mine safety management, the target area is usually those areas with a higher risk of gas outburst or that need to be monitored. This area can be a working face under excavation, an area that has been mined, or a part where other engineering operations are ongoing, as well as the entire mining site. The main purpose of selecting the target area is to improve the safety of the mine area by identifying and warning of possible coal and gas outburst incidents through detailed monitoring and analysis.

[0056] Physical information refers to various data that reflect the physical state of coal seams and rock formations in the target area. This information includes: Gas concentration: The gas content in the working face and the air around it is a key parameter for coal mine safety monitoring. Changes in gas concentration may indicate changes in the gas properties of coal seams, which in turn affect the risk of coal and gas outbursts; Pressure: The pressure data of coal seams and rock formations reflect the distribution and changes of ground stress. Changes in pressure are closely related to ground stress activity and are an important indicator for judging the risk of coal and gas outbursts; Microseismic signals: Vibration signals caused by ground stress activity recorded by microseismic sensors, including small events and large events. Microseismic signals can reveal the stress state of coal seams and rock formations and their changing process; Wind speed: The wind speed data in the tunnel reflects the ventilation status of the mining area. Changes in wind speed can affect the distribution of gas concentration and are an important parameter for evaluating the ventilation effect of mining areas.

[0057] Engineering operation information refers to the records and data of various engineering activities carried out in the target area. This information includes: Operation type: specific type of engineering activity, such as excavation, blasting, support, etc. Different types of operations have different effects on coal seams and rock formations, and need to be recorded and analyzed separately; Operation intensity: describes the intensity of engineering operations, such as the energy of blasting, the speed of excavation, etc. These parameters can affect the physical state of coal seams and gas release; Operation time: the time point and duration of specific operations. This helps to align physical information with engineering operation information in time and analyze the relationship between the two; Operation location: the specific location of the engineering operation in the mining area. This helps to spatially locate the source of changes in physical information.

[0058] By acquiring and analyzing the physical information and engineering operation information of the target area, we can fully understand the status and change trend of the mining area. This information can be combined to build a machine learning model for risk assessment and help generate subsequent early warning criteria to improve the safety management level of coal mines.

[0059] In the specific implementation, it is necessary to obtain the physical information and engineering operation information of the target area and preprocess this information to generate target features. This process includes three main steps: data collection, data preprocessing and feature extraction.

[0060] For data collection, a variety of sensors can be deployed in the target area to continuously monitor and record the physical information and engineering operation information of the mining area. Specifically, gas concentration sensors and wind speed sensors are used to monitor the gas concentration, ventilation status and gas outflow volume of the working face and its surroundings; pressure sensors are used to monitor the pressure changes of coal seams and rock formations; microseismic sensors record microseismic signals generated by the destruction of rock formations under the action of concentrated stress, including small events and large events; in addition, various engineering operation information is recorded, including the type, duration, operation intensity and location of operations such as excavation, blasting and support. These data provide a basis for subsequent analysis and feature extraction.

[0061] The small events and large events reflected by the microseismic signals recorded by the microseismic sensors will be described together in the small event analysis curve and large event analysis curve parts below.

[0062] In order to ensure data quality, the collected raw data needs to be preprocessed. First, data cleaning is performed to remove data during sensor failures and fill in missing values, and filtering technology is used to remove noise. Second, the physical information and engineering operation information are time-stamped to ensure that the two types of data can be analyzed synchronously.

[0063] As an optional implementation, the target feature includes: a first feature, a second feature, and a third feature; generating the target feature includes: extracting features based on the wavelet transform method for physical information to generate a first feature for characterizing mutation characteristics; extracting features based on the Fourier transform method to generate a second feature for characterizing periodic characteristics; generating a third feature for characterizing the dependency between the physical characteristics of the target area and the engineering activities based on the physical information and the engineering operation information;

[0064] In specific implementation, wavelet transform can be applied to analyze gas concentration, pressure and microseismic signals to identify mutation points and change trends. For example, wavelet transform can be used to extract the number of minor damages to soft rocks in the working face per minute, generate a small event curve to indicate the frequency of ground stress activity; wavelet transform can be used to extract the number of major damages to hard rocks in coal-bearing strata in the working face per minute, generate a large event curve to indicate the stress state of the hard roof rock layer; and by counting the energy released by small and large events per minute, an energy analysis curve is generated to indicate the energy of ground stress activity. Secondly, Fourier transform is applied to analyze the pressure signal, identify the main frequency components, generate periodic characteristic features, and indicate the amplitude and phase of different frequency components.

[0065] In a specific implementation, the preprocessed pressure signal data can be input into the Fourier transform algorithm. The Fourier transform converts the pressure signal in the time domain to the frequency domain, and analyzes the amplitude and phase of each frequency component in the signal. The main frequency components in the signal can be identified through the spectrum diagram, and these main frequency components reflect the characteristics of periodic changes in the pressure signal. For example, significant peaks of certain specific frequency components can be identified, and the frequencies corresponding to these peaks represent the periodic characteristics of ground stress activity. By analyzing the amplitude and phase of these frequency components, the contribution of different frequency components to the pressure signal can be further understood. The characteristic parameters of these frequency components (such as main frequency, amplitude, phase) are extracted to generate periodic characteristic features for subsequent risk assessment and early warning criterion generation. These periodic characteristic features can help identify the periodic laws in the process of coal and gas outbursts, and provide a basis for improving the accuracy and reliability of the early warning system.

[0066] For example, continuous microseismic signal data can be obtained from microseismic sensors installed near the working face. These signal data record the tiny vibrations generated when the soft rock of the working face is destroyed under the action of ground stress. Then, these microseismic signal data are preprocessed, including denoising, smoothing and standardization. Then, the preprocessed microseismic signal data are input into the wavelet transform algorithm. Wavelet transform can perform multi-scale analysis of signals in the time-frequency domain, thereby identifying mutation points and detailed features in the signal. By selecting appropriate wavelet functions and decomposition layers, wavelet transform can decompose microseismic signals into components of different frequencies and time scales.

[0067] In the decomposed signal, the micro-damage events that occur every minute can be identified. Specifically, by detecting the mutation points and local extreme points in the signal, the number of micro-damages per minute can be determined. The number of these micro-damage events is recorded in chronological order to generate a small event curve. The small event curve represents the frequency of micro-damages in the soft rock of the working face per minute, that is, the frequency of ground stress activity. For example, in a certain period of time, if there are more micro-damage events recorded per minute, it means that the ground stress activity of the working face in this period of time is more frequent, otherwise it means that the ground stress activity is less.

[0068] This feature extraction method based on wavelet transform can effectively capture the detailed changes in microseismic signals and improve the sensitivity and accuracy of the early warning system to ground stress activities.

[0069] Among them, the small event curve is composed of the number of tiny (small energy) damages that occur in the soft rock of the working face per minute, and it will only occur when the coal rock is damaged under the action of stress. When the working face stress and the coal rock strength are balanced (that is, when the stress intensity is less than the coal rock strength), no microseismic signal will be generated. The small event curve represents the frequency of ground stress activity, that is, whether there is ground stress activity; if small events occur throughout the day, especially when the working face is unmanned, it means that the concentrated stress of the working face is constantly shifting and intensifying, and it should be analyzed whether a geological structure zone is encountered in front of the working face. At this time, the working face has entered the "threat" area from the normal area. At this time, the changes in the occurrence conditions of the coal seams and the geological structure of the working face should be observed at all times.

[0070] The major event curve is composed of the number of (large energy) damages that occur to relatively hard rocks in the coal-bearing strata of the working face per minute. The distinction between major events and minor events can be made by calculating the energy of each event and distinguishing them according to the energy size. Specifically, an energy threshold is set. When the energy of the event exceeds the threshold, it is judged as a major event; otherwise, it is judged as a minor event. The setting of the energy threshold can be determined based on historical data and experimental results. In addition, major events and minor events differ in frequency domain characteristics. By analyzing the spectrum of the event signal, the amplitude and phase characteristics of different frequency components are identified. Major events have lower frequency components and higher energy; while minor events have higher frequency components and lower energy. Based on these frequency characteristics, major events and minor events can be further distinguished. The major event curve represents the strength and severity of rock damage under the action of ground stress activity on hard rocks. When major events begin to appear, it means that the stress intensity of the hard roof rock layer has reached a critical state. Monitoring personnel must pay close attention to changes in data and remind mine management personnel to slow down the excavation or mining speed and take appropriate safety measures. This means that the working face has entered the dangerous area of ​​outburst. When major events occur intensively and concentratedly, it means that the hard roof rock has been fractured on a large scale, and the energy accumulated in the fractured rock layer begins to impact the coal seam. The energy of the collected microseismic signal will mutate, and the energy value will reach tens of thousands of joules or more. The impact force on the coal seam will change the permeability of the coal seam, causing changes in the gas properties of the coal seam. Therefore, the major event curve is an important early warning curve for judging whether the working face has the danger of outburst.

[0071] In the trend analysis of major events, we always pay attention to the changes in the gas concentration, gas outburst volume and ventilation status curves in the same time and space. When the gas concentration and gas outburst volume curves suddenly become smaller after the intense activity of ground stress, it means that the pressure on the working face due to the action of ground stress is increasing rapidly, and the outstanding danger is entering a "dangerous" critical state. If the ground stress continues to increase, it is likely to cause fracture and damage to the hard rock formation on the top of the working face - resulting in a huge ground stress impact phenomenon. At this time, the working face should stop operations immediately to reduce the stress tension of the working face.

[0072] When the value of the energy curve increases sharply, it means that the coal seam at the working face has been impacted by the strong geostress caused by the fracture of the hard rock layer in the roof (excitation phenomenon has occurred). At this time, the gas concentration and gas outburst volume increase sharply under normal ventilation conditions, which means that the coal seam at the working face has been "excited" under the impact of geostress. Under the excitation, the adsorbed gas in the coal seam at the working face is gradually and rapidly desorbed and converted into free gas, and expansion energy that "drives" the outburst has been formed in the coal seam. This is an important precursor information of an impending outburst. At this time, the working face has entered a "critical" state of outburst danger warning, and safety protection measures such as power outage and evacuation of personnel in relevant areas should be immediately initiated.

[0073] As an optional implementation, generating a third feature based on physical information and engineering operation information to characterize the dependency between physical characteristics of a target area and engineering activities includes: aligning timestamps of the physical information and engineering operation information of the target area; determining target operation information based on the engineering operation information, and dividing time windows for each target operation information; determining physical feature change information based on the physical information of the target area within each time window; and generating the third feature based on the physical feature change information within each time window and the target operation information corresponding to each time window.

[0074] As an optional implementation, the third feature includes: a cross feature and a risk feature; based on the physical feature change information within each time window and the target operation information corresponding to each time window, generating the third feature includes: within each time window, based on the physical feature change information and the target operation information, creating the cross feature between the physical feature statistics and the target operation; within each time window, based on the physical feature change information and the target operation information, using the statistical threshold method to determine the potential risk information and generate the risk feature.

[0075] In the specific implementation, it is necessary to align the timestamps of the collected physical information and engineering operation information of the target area. The purpose of timestamp alignment is to ensure the consistency of physical information and engineering operation information in the time dimension so as to accurately analyze the relationship between the two.

[0076] For example, the timestamps of all physical information (such as gas concentration, pressure, microseismic signals, wind speed, etc.) and engineering operation information (such as excavation, blasting, support, etc.) are uniformly formatted. Use time synchronization technology to ensure that data collected by different sensors are recorded and processed under the same time reference. For data with incomplete timestamps, time alignment is performed through interpolation or other data completion methods to ensure that all data have a consistent time reference when analyzed.

[0077] In the specific implementation, the engineering operation information in the mining area is analyzed to determine the type, intensity, duration and location of each engineering operation. For each engineering operation, its start time and end time are determined.

[0078] A time window is divided for each target operation information. The length of the time window can be adjusted according to the specific operation type and the scope of influence. For example, for blasting operations, a shorter time window can be set, while for long-term excavation operations, a longer time window can be set.

[0079] Based on the physical information collected in each time window, calculate and analyze the change information of physical characteristics. For example, in each time window, count and analyze the changes in physical information such as gas concentration, pressure, microseismic signals, wind speed, etc. Perform differential analysis on the physical information and calculate the change rate and change amplitude of the physical characteristics. For example, calculate the change rate and change amplitude of gas concentration in the time window. Use signal processing technology (such as wavelet transform and Fourier transform) to extract the change characteristics of physical information and generate physical characteristic change information for subsequent analysis.

[0080] In each time window, the physical feature change information is correlated with the corresponding engineering operation information. Cross-features are created to reflect the dependency between physical feature changes and engineering operations. For example, by analyzing the temporal correspondence between gas concentration changes and blasting operations, cross-features representing the correlation between gas concentration and engineering operations are generated. Using statistical analysis methods, the statistics of each cross-feature, such as mean, standard deviation, correlation coefficient, etc., are calculated to further quantify the dependency between physical characteristics and engineering operations.

[0081] The analysis results in each time window are integrated to generate a complete set of third features that can accurately reflect the dependencies between the physical characteristics of the target area and the engineering activities.

[0082] In the specific implementation, the construction of cross features needs to be combined with specific parameter characteristics. According to the cross feature construction method, it can be classified into: basic operation cross features, statistical cross features, and advanced cross features;

[0083] Among them, basic operation cross-features refer to features constructed by addition, subtraction, multiplication and division operations between multiple features. For example, the product of the gas concentration change rate and the blasting operation intensity reflects the degree to which the gas concentration change is affected by the blasting intensity; the ratio of the gas concentration change rate to the operation duration reflects the change of gas concentration in unit time.

[0084] Statistical cross-features refer to features constructed by performing statistical operations between multiple features. For example, the statistics of physical features such as mean, standard deviation, maximum value, minimum value, etc. are calculated within a certain time window. In addition, the moving average and moving standard deviation of physical features can also be calculated to reflect the smooth changes and fluctuations of the data.

[0085] Advanced cross-features refer to features that can only be constructed by using certain technical means between multiple features, including: reducing the dimension of multiple physical features through principal component analysis (PCA), extracting the main components, and generating new features. For example, the principal components of the gas concentration change rate, pressure change rate, small event change rate, and large event change rate are used as new features; the mutual information between physical features and engineering operation features is calculated to reflect their dependency; the correlation between physical features and engineering operation features at different time lags is calculated using the time series correlation analysis method. For example, the correlation coefficient between the change in gas concentration within 1 minute after the blasting operation and the blasting operation is calculated.

[0086] In this way, a third feature can be systematically generated to characterize the dependency between the physical characteristics of the target area and engineering activities, providing a solid data foundation for subsequent risk assessment and early warning criterion generation. This method ensures the accuracy and consistency of the data and can deeply reveal the mutual influence relationship between physical information and engineering operation information.

[0087] Among them, the product characteristics directly reflect the interaction strength between physical characteristics and engineering operation characteristics. For example, the product characteristics of the gas concentration change rate and the blasting operation intensity can reveal the impact of blasting operations on gas concentration changes. This helps to identify the rapid changes in gas concentration caused by high-intensity operations, thereby timely warning of potential risks.

[0088] Ratio characteristics provide information about the relative changes of physical characteristics under different operating conditions. For example, the ratio of the gas concentration change rate to the operation duration can quantify the change in gas concentration per unit time and help identify the impact of the operation duration on the change in gas concentration. This is helpful in assessing the risk of gas accumulation caused by long-term operations.

[0089] The difference feature shows the absolute difference between physical features and engineering operation features. For example, the difference between the pressure change rate and the blasting operation intensity can reflect the inconsistency between the pressure change and the operation intensity, thereby identifying possible abnormal conditions, such as the risk caused by abnormally high pressure.

[0090] Sliding window statistics can smooth time series data and provide stable trend information. For example, the sliding mean and sliding standard deviation of the gas concentration change rate can reveal the changing trend and fluctuation of gas concentration in the short term, help identify emergencies and long-term changing trends, and adjust early warning strategies in a timely manner.

[0091] The moving standard deviation can reveal data fluctuations. For example, the moving mean and standard deviation of the pressure change rate can reflect the stability and volatility of pressure changes, help identify early signs of abnormal pressure changes, and improve the accuracy of early warnings.

[0092] PCA can reduce the dimensionality of multidimensional data, extract the main components, simplify the data structure and retain important information. For example, by extracting the main components of the gas concentration change rate, pressure change rate, small event change rate and large event change rate through PCA, comprehensive features can be generated to help identify the common changes between different physical features and improve the stability and prediction accuracy of the model.

[0093] Mutual information features can quantify the information exchange between physical features and engineering operation features and reveal the dependencies between them. For example, by calculating the mutual information between the gas concentration change rate and the blasting operation, strongly correlated feature pairs can be identified, improving the pertinence of feature selection and the effectiveness of the early warning model.

[0094] For example, a working face in a target area can be selected and divided into several time windows, for example, the length of the time window is set to 1 hour. In each time window, the collected physical characteristic change information includes the gas concentration change rate, pressure change rate, the number of small and large event changes of microseismic signals, and wind speed changes. Due to the great differences in coal seam occurrence conditions, data changes are jumpy and difficult to describe with an absolute number. It is more of a state or a state represented by a combination of multiple parameters.

[0095] In the specific implementation, the collected physical feature change information is correlated with the target operation information to generate cross-features. The specific steps are as follows: First, calculate the correlation between the gas concentration change rate and the blasting operation to generate the gas concentration change rate_blasting feature; calculate the correlation between the pressure change rate and the blasting operation to generate the pressure change rate_blasting feature; calculate the correlation between the small event and large event change rates and the blasting operation to generate the small event change rate_blasting feature and the large event change rate_blasting feature; calculate the correlation between the wind speed change and the blasting operation to generate the wind speed change_blasting feature.

[0096] Combine the physical feature change information with the operation intensity to generate cross features. For example, gas concentration change rate_intensity 3, pressure change rate_intensity 3, small event change rate_intensity 3, large event change rate_intensity 3, and wind speed change_intensity 3. In addition, combine the physical feature change information with the operation duration to generate features such as gas concentration change rate_30 minutes, pressure change rate_30 minutes, small event change rate_30 minutes, large event change rate_30 minutes, and wind speed change_30 minutes.

[0097] Among them, the above-mentioned strength 3 refers to the strength level. It can be understood that the cross-feature construction method here is only an example and does not constitute any form of specific cross-feature limitation.

[0098] Using statistical analysis methods, calculate the statistics of each cross-feature, such as mean, standard deviation, and correlation coefficient.

[0099] All generated cross-features are integrated to form a complete third feature set. For example, the final third feature set includes gas concentration change rate_blasting, gas concentration change rate_intensity3, gas concentration change rate_30 minutes, pressure change rate_blasting, pressure change rate_intensity3, pressure change rate_30 minutes, small event change rate_blasting, small event change rate_intensity3, small event change rate_30 minutes, large event change rate_blasting, large event change rate_intensity3, large event change rate_30 minutes, and wind speed change_blasting, wind speed change_intensity3, wind speed change_30 minutes and other features.

[0100] In a specific implementation, the third feature may also include a risk feature. The introduction of the risk feature can better help the subsequent model determine the risk situation of the current area. The introduction of the risk feature requires the setting of the risk threshold first. Each physical feature can correspond to a separate risk threshold, and the specific value of the risk threshold can be set based on historical data and domain knowledge.

[0101] Exemplarily, for a certain time window, if the gas concentration change rate exceeds the set high-risk threshold, it is marked as high risk and a risk feature is generated: gas concentration change rate_high risk; if the pressure change rate exceeds the set high-risk threshold, it is marked as high risk and a risk feature is generated: pressure change rate_high risk; if the change in the number of small events exceeds the set high-risk threshold, it is marked as high risk and a risk feature is generated: change in the number of small events_high risk; if the change in the number of large events exceeds the set high-risk threshold, it is marked as high risk and a risk feature is generated: change in the number of large events_high risk; if the wind speed change exceeds the set high-risk threshold, it is marked as high risk and a risk feature is generated: wind speed change_high risk.

[0102] Regarding S102 above:

[0103] In specific implementation, machine learning models are used to analyze and assess the risk of coal and gas outbursts. This process first involves collecting historical physical information and historical engineering operation information in the target area. These data may include but are not limited to physical parameters such as gas concentration, formation pressure, microseismic activity, wind speed, and information such as the type, intensity, time and location of engineering-related operations. These historical data provide training materials for machine learning models to help them learn and predict risk situations.

[0104] Specifically, the machine learning model used can be a support vector machine (SVM), decision tree, random forest, neural network, or any other algorithm suitable for handling time series prediction and classification problems. The model is selected based on its performance on similar problems, processing speed, ease of implementation, and interpretability.

[0105] The model will be trained based on the target features. Through supervised learning methods, the model learns to distinguish different risk levels and predict possible gas outburst events in the future.

[0106] After the model training is completed, it is used for risk assessment of real-time data. During operation, the system collects physical information and engineering operation information of the target area in real time. These real-time data are input into the trained model after the same preprocessing and feature extraction steps. The model will output a risk assessment result, which indicates the gas outburst risk level in the current monitoring area. The risk assessment information generated by the machine learning model provides dynamic and updated input for other optimization algorithms (such as whale optimization and non-dominated sorting genetic algorithm). This information helps these algorithms more effectively adjust their search and optimization strategies to adapt to the current mining conditions.

[0107] It should be noted that risk level analysis usually does not require high sensitivity to every detail and anomaly like predicting specific early warning criteria. It focuses more on identifying trends, patterns, and potential risks, which can make the model more robust to data fluctuations and incompleteness. When performing risk level analysis, the tasks of machine learning models are usually more generalized, such as classification or regression to a wider range of risk levels, rather than accurately predicting the timing and location of specific events. This simplifies the learning objectives of the model and reduces the risk of overfitting.

[0108] Regarding S103 above:

[0109] As an optional implementation, initializing the first iterative model based on the target feature includes: generating an initial population based on the target feature, wherein each individual represents an early warning criterion, wherein the early warning criterion includes at least one of the target features.

[0110] As an optional implementation, each individual also includes at least one characteristic parameter of the target characteristic included in the individual.

[0111] In the specific implementation, target features are extracted based on the pre-processed physical information and engineering operation information collected from the target area. These target features include the gas concentration change rate, pressure change rate, change in the number of small events, change in the number of large events, and wind speed change. Each individual represents an early warning criterion, which specifically includes a combination of multiple target features and a threshold or range for each feature. For example, an individual can be represented as "the gas concentration change rate is less than -0.05 and the change in the number of small events is greater than 5".

[0112] When generating the initial population, it is necessary to set the size of the population, for example, to 100 individuals. Initial individuals are randomly generated based on possible combinations of target features. For example, the criterion combination for individual 1 is that the gas concentration change rate is less than -0.04 and the pressure change rate is greater than 0.1, and the criterion combination for individual 2 is that the number of small events changes is greater than 3 and the number of large events changes is greater than 2. Each individual consists of a randomly selected target feature and its corresponding random threshold.

[0113] To ensure the diversity of the population, randomly generated individuals should cover a wide range of feature combinations and thresholds. Specific generation rules or restrictions can be set to prevent individuals from being too similar. Each individual is stored in the population data structure as a combination of early warning criteria.

[0114] In this way, the system can initialize the first iteration model based on the target features and generate an initial population. Each individual represents a potential combination of early warning criteria, covering at least one of the target features. This initialization method ensures the diversity and wide coverage of the population, providing a basis for the subsequent iterative optimization process. The system will construct an initial population by randomly generating individuals and ensuring diversity, so that it can effectively optimize early warning criteria in a complex and changeable mining environment.

[0115] As an optional implementation, the first iterative model includes: a whale optimization sub-model and a non-dominated sorting genetic sub-model; the iterative search for target warning criterion combination includes:

[0116] Based on the physical information and engineering operation information of the target area, a fitness function is constructed, wherein the fitness function is used to evaluate at least one of the following performance indicators: warning accuracy, false alarm rate, response speed, and computing cost;

[0117] In each iteration cycle of the first iteration model, the whale optimization sub-model is used to update the position of each individual, wherein the position update of each individual is used to update the characteristic parameters that characterize the target characteristics;

[0118] The non-dominated sorting genetic sub-model is used to perform non-dominated sorting on the individuals updated by the whale optimization sub-model, determine the dominance level of each individual, and calculate the crowding distance of each individual; based on the dominance level of each individual and the crowding distance of each individual, select the parent individuals and perform crossover and mutation operations to generate new offspring individuals;

[0119] In the specific implementation, the whale optimization sub-model is constructed based on the whale optimization algorithm (Whale Optimization Algorithm, WOA) as the main body, and the non-dominated sorting genetic sub-model is constructed based on the non-dominated sorting genetic algorithm II (Non-dominated Sorting Genetic Algorithm II, NSGA-II) as the main body.

[0120] Among them, WOA excels in global search, i.e. exploration, which enables the algorithm to quickly identify potentially favorable areas in the entire search space. It effectively explores a wide range of solution spaces by simulating the hunting behavior of whales, thereby avoiding early convergence to local optimal solutions.

[0121] NSGA-II is good at exploitation, especially in multi-objective optimization. It ensures the diversity of solutions through non-dominated sorting and crowding distance calculation, fine-tunes the solutions in the already explored favorable areas, and improves the quality of the solutions.

[0122] The multi-objective optimization capability of NSGA-II allows multiple key performance indicators to be considered simultaneously, such as the accuracy of warnings, false alarm rate, and response speed. This is particularly important for building an efficient early warning system, because in practical applications, it is necessary to find the best balance between these indicators.

[0123] Combining WOA ensures that the optimal solution is found globally, while NSGA-II precisely adjusts these solutions to meet the optimization of multiple objectives.

[0124] Individual optimization algorithms perform poorly on certain types of problems, especially when the solution space is complex or the problem is variable. Combining the two algorithms can enhance the robustness of the system, allowing the early warning system to maintain good performance in different mining environments and under different working conditions.

[0125] Directly using machine learning models to find early warning criteria may ignore the multi-objective nature of the problem, that is, machine learning models usually optimize a single objective (such as minimizing error). Using a combination of WOA and NSGA-II can not only explore multiple possible early warning criteria, but also find the best trade-off between various performance indicators, thereby generating more comprehensive and adaptable early warning criteria.

[0126] Furthermore, the optimization algorithm can be performed without explicit labels, which is particularly useful for processing complex mining data (which may be insufficiently labeled).

[0127] In this way, the system not only discovers effective early warning criteria, but also ensures that these criteria perform well across multiple dimensions.

[0128] As an optional implementation manner, in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stopping iteration and outputting the current early warning criterion as a target early warning criterion includes:

[0129] In response to the generation of a new offspring individual, the fitness of the new offspring individual is calculated based on the fitness function;

[0130] In response to the fitness of the new offspring individual being greater than the target threshold, the iteration is stopped and the current warning criterion is output as the target warning criterion; otherwise, the new offspring individual is combined with the initial population as the population of the first iterative model in the next round of iteration.

[0131] In the specific implementation, a fitness function is constructed based on the physical information and engineering operation information of the target area. The fitness function is used to evaluate the performance of the early warning criterion combination, including the following indicators: Early warning accuracy: the accuracy of the system in predicting coal and gas emergencies, which can be evaluated by comparing the prediction results with the actual occurrence of the event; False alarm rate: the frequency of the system issuing false alarms when there are no emergencies, which can be calculated by counting the ratio of the number of false alarms to the total number of alarms; Response speed: the time interval from data input to early warning output of the system, reflecting the real-time nature of the system, which can be evaluated by measuring the processing time of the early warning system; Computational cost: the computing resource consumption of the algorithm during operation, including time and memory, which can be measured by monitoring the system resource usage.

[0132] These performance indicators can be set based on historical data and each actual work requirement, and used to evaluate the fitness of each early warning criterion combination.

[0133] In a specific implementation, in each iteration cycle of the first iteration model, the whale optimization submodel (WOA) is first used to update the position of each individual. The position of each individual in the initial population is randomly generated to represent different early warning criterion combinations.

[0134] WOA updates the individual's position by simulating the whale's hunting behavior. Specifically:

[0135] Contraction and Surrounding Mechanism: Simulates the behavior of a whale approaching its prey, and improves the accuracy of the search by gradually reducing the search area.

[0136] Spiral update path: simulates the behavior of whales spiraling towards their prey, and explores new areas in the solution space by performing spiral searches in local areas. For example, the position of the individual is slightly adjusted on the spiral path, and the gas concentration change rate is updated to -0.045 and the pressure change rate is 0.11.

[0137] Random search: By randomly selecting individuals in the population and moving them to explore different areas in the solution space, we can avoid falling into the local optimum. For example, randomly select the positions of other individuals to update, the gas concentration change rate is -0.05, and the pressure change rate is 0.13.

[0138] The position update of each individual is used to update the characteristic parameters that characterize the target characteristics. For example, the gas concentration change rate of an individual is updated from -0.05 to -0.04, and the pressure change rate is updated from 0.1 to 0.15.

[0139] After the whale optimization submodel updates the individual positions, the non-dominated sorting genetic submodel (NSGA-II) is used to perform non-dominated sorting on the updated individuals according to the fitness function to determine the dominance level of each individual. Non-dominated sorting divides individuals into different levels, and individuals with lower levels are better than those with higher levels. For example, individuals with a gas concentration change rate of -0.04 and a pressure change rate of 0.12 are ranked in a lower level, while individuals with a gas concentration change rate of -0.045 and a pressure change rate of 0.11 are ranked in a higher level.

[0140] Calculate the crowding distance of each individual and evaluate the distribution of individuals in the solution space. Crowding distance is used to maintain the diversity of the population and prevent the algorithm from converging to the local optimum too early. For example, calculate the distance between each individual and its neighbor in the fitness space. The larger the distance, the sparser the individual is, and it is preferred to be retained.

[0141] Select parent individuals from the population based on dominance rank and crowding distance. Give priority to individuals with lower dominance rank and larger crowding distance. For example, select individuals with gas concentration change rate -0.04 and pressure change rate 0.12 as parents.

[0142] Perform crossover and mutation operations on the selected parent individuals to generate new offspring individuals. The crossover operation generates new individuals by exchanging some genes of the parent individuals, and the mutation operation introduces diversity by randomly changing some genes of the individuals. For example, the gas concentration change rate and pressure change rate of two parent individuals are crossed to generate new offspring individuals, and the gas concentration change rate is updated to -0.045 and the pressure change rate is updated to 0.125.

[0143] The above iteration cycle is repeated until any of the following convergence conditions are met: 1. The maximum number of iterations is reached. For example, the maximum number of iterations is set to 100. 2. The fitness of the current warning criterion combination is greater than the target threshold. For example, the fitness target threshold is set to 0.9.

[0144] At the end of each iteration, the new offspring individuals are combined with the initial population as the population in the next iteration. For example, the new offspring individuals with a gas concentration change rate of -0.045 and a pressure change rate of 0.125 are combined with other individuals in the initial population to form a new population. If the fitness of the new offspring individuals is greater than the preset target threshold, the iteration is stopped and the current warning criterion is output as the target warning criterion. For example, when the fitness of the new offspring individuals reaches 0.92, the iteration is stopped and it is used as the target warning criterion.

[0145] It is understandable that the principles of WOA and NSGA-II algorithms are public content, so this disclosure does not elaborate on their principle analysis. For the specific construction and use of the two, please refer to the public content.

[0146] In this way, combining the global search capability of the whale optimization sub-model and the multi-objective optimization characteristics of the non-dominated sorting genetic sub-model, the optimal combination of coal and gas outburst warning criteria can be efficiently found and verified, ensuring that the early warning system has high accuracy, low false alarm rate and rapid response capabilities in complex and changeable mining environments.

[0147] For example, when the population is initialized, 50 early warning criteria combination individuals are created, each of which includes the gas concentration change rate, pressure change rate and cross-features associated with the blasting operation. The characteristic parameters of each individual are randomly set, for example: Individual A: gas concentration change rate threshold> 0.05, pressure change rate threshold> 0.03, blasting intensity and gas concentration change rate cross-feature coefficient> 2; Individual B: gas concentration change rate threshold> 0.03, pressure change rate threshold> 0.04, blasting intensity and gas concentration change rate cross-feature coefficient> 1.5;

[0148] The fitness function is calculated based on the warning accuracy, false alarm rate, response speed and computational cost. The third feature (cross feature) can be set with a high weight in the fitness evaluation because it can effectively reflect the correlation between engineering operations and gas risks.

[0149] In actual evaluation, if an individual can accurately predict the sudden increase in gas concentration associated with blasting operations, its fitness score will be higher.

[0150] Whale Optimization Algorithm (WOA) operation: Based on the relative performance of individuals, the position is updated. For example, if individual A accurately predicts three sudden increases in gas concentration in the simulation scenario, while individual B only predicts one, individual A will be used as the guide individual. Other individuals will move closer to the characteristic parameters of individual A, such as increasing the cross-characteristic coefficient of blast intensity and gas concentration change rate.

[0151] Application of Non-dominated Sorting Genetic Algorithm (NSGA-II): At the end of each iteration cycle, the updated population is subjected to non-dominated sorting and crowding distance calculation. Individuals with high fitness are selected for crossover and mutation operations to generate new individuals. For example, individuals A and C are selected for crossover to generate new individual D with mixed characteristic parameters.

[0152] Exemplarily, during the iteration process, the fitness of all individuals in the population (such as individuals A, B, etc.) is evaluated. Assume that individual A has the highest fitness, so it becomes the leading individual. Other individuals will update their positions to move closer to the leading individual. For example, individual B may update its gas concentration change rate threshold and pressure change rate threshold to make them closer to the values ​​of individual A. Position updates also include random exploration components to avoid local optimal solutions, such as randomly adjusting the coefficients of the cross-features of blasting intensity and gas concentration change rate.

[0153] After executing WOA, individuals are sorted according to fitness. Individual A may be at the top of the sort due to its high fitness. The crowding distance of each individual (such as individuals A and B) is calculated, and the parent individuals are selected for crossover and mutation based on the non-dominated sorting and crowding distance. For example, genes may be selected from individuals A and B to generate a new offspring individual C. Individual C may inherit the gas concentration change rate threshold of A and the pressure change rate threshold of B.

[0154] The iteration continues until the preset maximum number of iterations is reached or the fitness of the newly generated offspring individual (such as individual C) exceeds the preset target threshold. At termination, the individual with the highest fitness, such as individual C, is selected as the final warning criterion combination output. In each iteration, the characteristic parameters of the individual (such as gas concentration change rate threshold, pressure change rate threshold, and cross-characteristic coefficient) are dynamically adjusted according to the exploration of WOA and the optimization of NSGA-II. This dynamic adjustment helps the system adapt to the complex and changeable mining environment and improves the responsiveness and accuracy of the early warning system. By combining the two optimization strategies, the system can effectively balance global search and local refinement to ensure the comprehensiveness and practicality of the early warning criteria.

[0155] Among them, the fitness evaluation of individuals can be carried out in a simulation scenario. For example, the early warning criteria represented by the individual are input into the machine learning model responsible for the prediction of coal and gas outbursts in the mining area. The machine learning model will predict the outburst situation on the historical data set of the mining area based on the feature combination represented by the individual. The fitness of the individual can be obtained by comparing the predicted results with the actual results in the historical data set.

[0156] It should be emphasized that the output information of the present disclosure is an early warning criterion, that is, a certain or several feature combinations related to the mining area. Based on these early warning criteria, predictions can be made through machine learning, data monitoring, etc. For example, the present disclosure finally confirmed three early warning criterion combinations, which can be used as input features of a certain prediction model, or the weights of these three feature combinations in the prediction model can be increased, or the features corresponding to these feature combinations can be directly monitored through sensors.

[0157] As an optional implementation manner, adjusting the model parameters of the first iterative model based on the risk assessment information includes:

[0158] Based on the risk assessment information, adjusting population information, search intensity parameters, spiral update parameters, and random search probability in the whale optimization sub-model;

[0159] Based on the risk assessment information, the population information, crossover rate, and mutation rate in the non-dominated sorting genetic submodel are adjusted.

[0160] In specific implementation, the risk assessment information reflects the current safety status of the mining area, including the analysis results of key indicators such as gas concentration, pressure changes, and microseismic activity. This information is used to dynamically adjust the calculation parameters of the two sub-models, so as to more accurately search and optimize the early warning criteria.

[0161] Depending on the outcome of the risk assessment, it may be necessary to increase or decrease the size of the search population. For example, if the risk level increases, a more extensive search may be necessary to ensure that no potential high-risk areas are missed, in which case the population size may be increased.

[0162] Search intensity parameter: adjusts the search intensity to reflect the response to the current risk level. A high risk assessment may prompt the algorithm to increase local search capabilities to explore possible high-risk solutions in more detail.

[0163] Spiral Update Parameter: This parameter determines the tightness of the search spiral during the spiral update process that simulates whale hunting behavior. Depending on the risk assessment, the tightness of the spiral can be adjusted to strengthen or weaken the balance between global and local search.

[0164] Random search probability: Adjust the proportion of random search to cope with different risk conditions. In cases of high uncertainty, more randomness is needed to explore unknown areas of possible high risk.

[0165] For the non-dominated sorting genetic submodel, similarly, the population size is adjusted to adapt to different risk assessment results to ensure that the multi-objective optimization can cover all key risk factors. In the case of higher risk, the crossover rate needs to be increased to quickly generate possible excellent solutions. In high-risk situations, by increasing the mutation rate, more randomness and innovation can be introduced to find potential solutions that are not covered by the current population.

[0166] In this way, the iterative model can respond more flexibly to the actual safety conditions in the mining area, effectively improving the accuracy and adaptability of the early warning system. This strategy enables the model to not only work well under stable conditions, but also quickly adapt and provide reliable early warnings in a changing real environment.

[0167] Based on the same inventive concept, the embodiment of the present disclosure also provides a system for automatically finding and verifying coal and gas outburst warning criteria corresponding to the method for automatically finding and verifying coal and gas outburst warning criteria. Since the principle of solving the problem by the system in the embodiment of the present disclosure is similar to the method for automatically finding and verifying coal and gas outburst warning criteria in the above-mentioned embodiment of the present disclosure, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0168] Reference Figure 3 FIG. 1 is a schematic diagram of a system for automatically finding and verifying coal and gas outburst early warning criteria provided by an embodiment of the present disclosure, the system comprising:

[0169] The acquisition unit 10 is used to obtain physical information and engineering operation information of the target area, and perform preprocessing to generate target features;

[0170] An evaluation unit 20, configured to perform risk evaluation based on the target features using a machine learning model to generate risk evaluation information; wherein the machine learning model is constructed based on historical physical information and historical engineering operation information of the target area;

[0171] The iteration unit 30 is used to initialize a first iterative model based on the target feature; adjust the model parameters of the first iterative model based on the risk assessment information, and iteratively search for a target warning criterion; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the warning criterion during the iteration process being greater than a target threshold, stop the iteration and output the current warning criterion as the target warning criterion.

[0172] The present disclosure also provides an electronic device, such as Figure 4FIG. 1 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure, including:

[0173] A processor 31 and a memory 32; the memory 32 stores machine-readable instructions executable by the processor 31, and the processor 31 is used to execute the machine-readable instructions stored in the memory 32. When the machine-readable instructions are executed by the processor 31, the processor 31 performs the following steps:

[0174] Obtain physical information and engineering operation information of the target area, perform preprocessing, and generate target features;

[0175] Based on the target characteristics, a machine learning model is used to perform risk assessment to generate risk assessment information; wherein the machine learning model is constructed based on historical physical data and historical operation information of the target area;

[0176] Based on the target features, a first iterative model is initialized; based on the risk assessment information, model parameters of the first iterative model are adjusted, and a target early warning criterion is iteratively searched; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process is greater than a target threshold, the iteration is stopped and the current early warning criterion is output as the target early warning criterion.

[0177] The above-mentioned memory 32 includes internal memory 321 and external memory 322; the memory 321 here is also called internal memory, which is used to temporarily store the calculation data in the processor 31, as well as the data exchanged with the external memory 322 such as the hard disk. The processor 31 exchanges data with the external memory 322 through the internal memory 321.

[0178] The specific execution process of the above instructions can refer to the steps of the method for automatically finding and verifying coal and gas outburst warning criteria described in the embodiment of the present disclosure, which will not be repeated here.

[0179] The embodiment of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for automatically finding and verifying coal and gas outburst early warning criteria described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0180] The embodiments of the present disclosure also provide a computer program product, which carries a program code. The instructions included in the program code can be used to execute the steps of the method for automatically finding and verifying coal and gas outburst warning criteria described in the above method embodiment. Please refer to the above method embodiment for details, which will not be repeated here.

[0181] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0182] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

[0183] It should be understood that determining B based on A does not mean determining B only based on A. B can also be determined based on A and / or other information.

[0184] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0185] In the description of this specification, the description with reference to the terms "in a specific implementation", "exemplary", "for example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0186] The preferred embodiments of the present invention disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details in detail, nor do they limit the present application to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can understand and use the present application well. The present application is limited only by the claims and their full scope and equivalents.

Claims

1. A method for automatically finding and verifying coal and gas outburst early warning criteria, characterized in that: include: Obtain physical information and engineering operation information of the target area, perform preprocessing, and generate target features; Based on the target characteristics, a machine learning model is used to perform risk assessment to generate risk assessment information; wherein the machine learning model is constructed based on historical physical information and historical engineering operation information of the target area; Initializing a first iterative model based on the target feature; adjusting model parameters of the first iterative model based on the risk assessment information, and iteratively searching for a target early warning criterion; in response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stopping iteration and outputting the current early warning criterion as the target early warning criterion; The target features include: a first feature, a second feature, and a third feature; the generated target features include: For the physical information, feature extraction is performed based on the wavelet transform method to generate a first feature for characterizing the mutation characteristic; feature extraction is performed based on the Fourier transform method to generate a second feature for characterizing the periodic characteristic; Based on the physical information and the engineering operation information, a third feature is generated to characterize the dependency between the physical characteristics of the target area and the engineering activities.

2. The method according to claim 1, characterized in that The generating, based on the physical information and the engineering operation information, a third feature for characterizing the dependency relationship between the physical characteristics of the target area and the engineering activities comprises: Performing time stamp alignment on the physical information and engineering operation information of the target area; Based on the engineering operation information, determine target operation information, and divide a time window for each target operation information; Determining physical characteristic change information based on the physical information of the target area within each of the time windows; The third feature is generated based on the physical feature change information within each of the time windows and the target operation information corresponding to each of the time windows.

3. The method according to claim 2, characterized in that The third feature includes: an intersection feature and a risk feature; the third feature is generated based on the physical feature change information in each time window and the target operation information corresponding to each time window, including: In each of the time windows, based on the physical feature change information and the target job information, create the cross-feature between the physical feature statistics and the target job; In each of the time windows, based on the physical feature change information and the target operation information, a statistical threshold method is used to determine potential risk information and generate the risk feature.

4. The method according to claim 1, characterized in that: Initializing the first iterative model based on the target feature includes: Based on the target features, an initial population is generated, wherein each individual represents an early warning criterion, wherein the early warning criterion includes at least one of the target features.

5. The method according to claim 4, characterized in that Each of the individuals also includes at least one characteristic parameter of the target characteristic included in the individual.

6. The method according to any one of claims 1 to 5, characterized in that: The first iterative model includes: a whale optimization sub-model and a non-dominated sorting genetic sub-model; the iterative search for target warning criterion combination includes: Based on the physical information and engineering operation information of the target area, construct a fitness function, wherein the fitness function is used to evaluate at least one of the following performance indicators: warning accuracy, false alarm rate, response speed, and computing cost; In each iteration cycle of the first iteration model, the whale optimization sub-model is used to update the position of each of the individuals, wherein the position update of each of the individuals is used to update the characteristic parameters that characterize the target characteristics; The non-dominated sorting genetic sub-model is used to perform non-dominated sorting on the individuals updated by the whale optimization sub-model, determine the dominance level of each individual, and calculate the crowding distance of each individual; based on the dominance level of each individual and the crowding distance of each individual, select the parent individual and perform crossover and mutation operations to generate new offspring individuals.

7. The method according to claim 6, characterized in that In response to the first iterative model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stopping iteration and outputting the current early warning criterion as a target early warning criterion includes: In response to the generation of a new offspring individual, the fitness of the new offspring individual is calculated based on the fitness function; In response to the fitness of the new offspring individual being greater than the target threshold, the iteration is stopped and the current warning criterion is output as the target warning criterion; otherwise, the new offspring individual is combined with the initial population as the population of the first iterative model in the next round of iteration.

8. The method according to claim 7, characterized in that The adjusting the model parameters of the first iterative model based on the risk assessment information includes: Based on the risk assessment information, adjusting population information, search intensity parameters, spiral update parameters, and random search probability in the whale optimization sub-model; Based on the risk assessment information, the population information, crossover rate, and mutation rate in the non-dominated sorting genetic submodel are adjusted.

9. A system for automatically finding and verifying coal and gas outburst early warning criteria, characterized in that: The method for automatically finding and verifying coal and gas outburst early warning criteria as described in any one of claims 1 to 8 comprises: The acquisition unit is used to obtain the physical information and engineering operation information of the target area, and perform preprocessing to generate target features; An evaluation unit, configured to perform risk evaluation based on the target characteristics using a machine learning model to generate risk evaluation information; wherein the machine learning model is constructed based on historical physical information and historical engineering information of the target area; An iteration unit, configured to initialize a first iteration model based on the target feature; adjust a model parameter of the first iteration model based on the risk assessment information, and iteratively search for a target early warning criterion; in response to the first iteration model reaching a maximum number of iterations, or the fitness of the early warning criterion during the iteration process being greater than a target threshold, stop iteration and output the current early warning criterion as the target early warning criterion; The target features include: a first feature, a second feature, and a third feature; the generated target features include: For the physical information, feature extraction is performed based on the wavelet transform method to generate a first feature for characterizing the mutation characteristic; feature extraction is performed based on the Fourier transform method to generate a second feature for characterizing the periodic characteristic; Based on the physical information and the engineering operation information, a third feature is generated to characterize the dependency between the physical characteristics of the target area and the engineering activities.

Images