Artificial Intelligence-Based Intelligent Prediction Method and Early Warning System for Coal Mine Water Hazards

Abstract

This application provides an intelligent prediction method and early warning system for coal mine water hazards based on artificial intelligence, which relates to coal mine safety technology. In this solution, the computer device inputs the old kiln water monitoring data within the first preset time period before the current moment uploaded by the sensing layer device into the old kiln water hazard prediction model for prediction processing, and obtains the predicted data of the water inflow in the working face within the second preset time period after the current moment to determine whether an old kiln water hazard accident will occur. By applying this method, it is possible to achieve high-precision prediction of old kiln water hazard accidents in the coal mine area to be predicted, thereby improving the safety level of the coal mine.

Description

Technical Field

[0001] This application relates to coal mine safety technology, and in particular, to an intelligent prediction method and early warning system for coal mine water hazards based on artificial intelligence. Background Art

[0002] With the continuous development of artificial intelligence technology, the intelligent construction of coal mines has gradually emerged. However, coal mine water hazard accidents occur frequently.

[0003] To solve the above problems, coal mine water hazard detection technologies have been gradually applied. Among them, neural network models have been widely introduced into the design and application of coal mine water hazard detection technologies. Specifically, according to the selected type of neural network model, the structure of the neural network model is constructed manually, and the hyperparameters therein are calibrated to determine the initial neural network model. Then, a large amount of coal mine water hazard data is learned and analyzed to obtain a neural network model for predicting coal mine water hazard data.

[0004] However, there are still certain errors in the above coal mine water hazard detection technologies when using neural network models to predict coal mine water hazard data, which are likely to lead to the occurrence of coal mine water hazard accidents. Summary of the Invention

[0005] This application provides an intelligent prediction method and early warning system for coal mine water hazards based on artificial intelligence to solve the problem of low prediction accuracy of coal mine water hazard data in the prior art.

[0006] In a first aspect, this application provides an intelligent prediction method for coal mine water hazards based on artificial intelligence. The method includes:

[0007] Receiving the old kiln water monitoring data of the to-be-predicted area of the coal mine uploaded by the sensing layer device within a first preset time period before the current moment, where the old kiln water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old kiln damage monitoring data;

[0008] Inputting the old kiln water monitoring data of the to-be-predicted area within the first preset time period into an old kiln water hazard prediction model for prediction processing to obtain the predicted working face water inrush volume data within a second preset time period after the current moment, where the old kiln water hazard prediction model is a long short-term memory neural network model LSTM for predicting the working face water inrush volume that is pre-optimized based on the genetic algorithm GA, and the GA is used to determine the parameters of the LSTM model; the length of the first preset time period is greater than the length of the second preset time period;

[0009] Determining whether an old kiln water hazard accident will occur according to the predicted working face water inrush volume data of the to-be-predicted area within the second preset time period.

[0010] In a possible design of the first aspect, the process of obtaining the old kiln water disaster prediction model includes:

[0011] According to the GA, determine the target parameter configuration scheme of the LSTM model;

[0012] According to the target parameter configuration scheme, configure the target parameters in the LSTM model to obtain the old kiln water disaster prediction model.

[0013] In a possible design of the first aspect, the determining the target parameter configuration scheme of the LSTM model according to the GA includes:

[0014] Step a: Obtain a sample data set based on historical old kiln water monitoring data. The historical old kiln water monitoring data includes all historical water source monitoring data, historical working face water inrush monitoring data, and historical old kiln water damage monitoring data. The sample data set includes multiple groups of data samples. Each group of data samples includes old kiln water monitoring data for two consecutive time periods, and the duration of the first time period is greater than or equal to the duration of the first preset time period, and the duration of the second time period is the same as the duration of the second preset time period;

[0015] Step b: Initialize N groups of first parameter configuration schemes, selection operator, adaptive crossover probability, adaptive mutation probability, and preset maximum iteration times of the GA, where N is a positive integer. Each group of first parameter configuration schemes includes the number of hidden layer neurons, learning rate, preset iteration times, minimum batch size, regularization parameter, weight parameter, and bias term parameter;

[0016] Step c: Construct N LSTM models with exactly the same structure to obtain N first LSTM models;

[0017] Step d: Perform chromosome coding on each group of first parameter configuration schemes to obtain the corresponding first chromosome;

[0018] Step e: For each first LSTM model, select a first chromosome corresponding to the first parameter configuration scheme, and configure the initial parameters in the first LSTM model to obtain N second LSTM models, where the initial parameters of the first LSTM model include the initial number of hidden layer neurons, initial learning rate, initial preset iteration times, initial minimum batch size, initial regularization parameter, initial weight parameter, and initial bias term parameter;

[0019] Step f: Select N groups of sample data from the sample data set, and input the historical old kiln water monitoring data in the first time period of the N groups of sample data into the N second LSTM models for prediction to obtain the predicted data of the water inrush volume of the working face in the second time period;

[0020] Step g: Calculate the fitness value of each first chromosome according to the predicted water inflow data of the working face in the N second time periods and the preset fitness function of the GA.

[0021] Step h: Sort the fitness values of the N first chromosomes from high to low according to the fitness value of each first chromosome, obtain the number of the first chromosome corresponding to the highest fitness value, and record it as the target chromosome number.

[0022] Step i: Screen the first chromosomes according to the fitness value of each first chromosome and the selection operator to obtain N screened first chromosomes.

[0023] Step j: Calculate the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome according to the fitness value of each screened first chromosome.

[0024] Step k: Based on the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome, perform adaptive crossover and mutation operations on the screened first chromosomes to generate N second chromosomes.

[0025] Step l: Determine whether the current iteration number reaches the preset maximum iteration number. If it does not reach the preset maximum iteration number, determine the N second chromosomes as the new N first chromosomes, and repeat steps e to k until the preset maximum iteration number is reached. When the preset maximum iteration number is reached, determine the first parameter configuration scheme corresponding to the obtained target chromosome number as the target parameter configuration scheme.

[0026] In a possible design of the first aspect, the screening process of the first chromosomes according to the fitness value of each first chromosome and the selection operator to obtain N screened first chromosomes includes:

[0027] Determine the first chromosomes with fitness values satisfying the preset fitness condition as the first part of chromosomes from all the first chromosomes.

[0028] Determine the second part of chromosomes from all the first chromosomes according to the roulette wheel method. The N screened first chromosomes include the first part of chromosomes and the second part of chromosomes.

[0029] In a possible design of the first aspect, the fitness value of each first chromosome is calculated using the following formula:

[0030]

[0031] Where represents the fitness value of the nth first chromosome, f MAE (n) represents the mean absolute error value of the nth first chromosome, f RMSE (n) represents the root mean square error value of the nth first chromosome.

[0032] In a possible design of the first aspect, according to the fitness value of each screened first chromosome, calculate the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome, including:

[0033] For each screened first chromosome, determine the adaptive crossover probability of the screened first chromosome according to a preset initial crossover probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome;

[0034] Determine the adaptive mutation probability of the screened first chromosome according to a preset initial mutation probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome.

[0035] In a possible design of the first aspect, based on the predicted water inflow data of the working face in the second preset time period in the area to be predicted, determine whether an old kiln water hazard accident will occur. The method includes:

[0036] If the predicted water inflow data of the working face in the second preset time period is lower than the preset water inflow data of the working face, it is determined that no old kiln water hazard accident will occur in the area during the second preset time period;

[0037] If the predicted water inflow data of the working face in the second preset time period is higher than or equal to the preset water inflow data of the working face, it is determined that an old kiln water hazard accident will occur in the area during the second preset time period.

[0038] In a possible design of the first aspect, the method further includes:

[0039] If it is determined that no old kiln water hazard accident will occur in the area during the second preset time period, push a safety notice message to the management personnel, where the safety notice message is used to indicate that no old kiln water hazard accident will occur in the area;

[0040] If it is determined that an old kiln water hazard accident will occur in the area during the second preset time period, push a safety warning message to the management personnel, where the safety warning message is used to indicate that an old kiln water hazard accident will occur in the area.

[0041] In a second aspect, the present application provides a computer device, including: a receiver, a processor, and a memory connected to the processor;

[0042] The memory stores computer-executable instructions;

[0043] The processor executes the computer-executable instructions stored in the memory to implement the artificial intelligence-based intelligent prediction method for coal mine water hazards as described in any one of the first aspects.

[0044] In a third aspect, the present application provides an early warning system for coal mine water hazards based on artificial intelligence, including: a computer device and a sensing layer device connected to the computer device;

[0045] The sensing layer device includes: a water source monitoring sensor, a working face water inrush monitoring sensor, and an old kiln damage depth sensor;

[0046] The water source monitoring sensor is used to monitor the old kiln water source data;

[0047] The working face water inrush monitoring sensor is used to monitor the old kiln water working face water inrush data;

[0048] The old kiln damage depth sensor is used to monitor the old kiln damage depth data;

[0049] The computer device is used to execute the artificial intelligence-based intelligent prediction method for coal mine water hazards as described in any one of the first aspects.

[0050] In a fourth aspect, the present application provides an artificial intelligence-based intelligent prediction device for coal mine water hazards, including:

[0051] A receiving module, configured to receive the old kiln water monitoring data of the to-be-predicted area of the coal mine uploaded by the sensing layer device within a first preset time period before the current moment, where the old kiln water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old kiln damage monitoring data;

[0052] A processing module, configured to input the old kiln water monitoring data of the to-be-predicted area within the first preset time period into an old kiln water hazard prediction model for prediction processing to obtain the predicted working face water inrush volume data within a second preset time period after the current moment, where the old kiln water hazard prediction model is a long short-term memory neural network model LSTM for predicting the working face water inrush volume that is pre-optimized based on the genetic algorithm GA, and the GA is used to determine the parameters of the LSTM model; the length of the first preset time period is greater than the length of the second preset time period;

[0053] The processing module is further configured to determine whether an old kiln water hazard accident will occur according to the predicted working face water inrush volume data of the to-be-predicted area within the second preset time period.

[0054] Fifth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the artificial intelligence-based intelligent prediction method for coal mine water hazards as described in any one of the first aspect.

[0055] Sixth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the artificial intelligence-based intelligent prediction method for coal mine water hazards as described in any one of the first aspect.

[0056] The artificial intelligence-based intelligent prediction method and early warning system for coal mine water hazards provided by the present application relate to coal mine safety technology. In the area to be predicted in the coal mine, the sensing layer devices collect the old kiln water monitoring data within the first preset time period before the current moment and transmit it to the computer device to ensure the real-time nature of the monitoring data. The computer device then uses the old kiln water hazard prediction model to obtain the predicted data of the water inflow in the working face within the second preset time period after the current moment, and judges whether an old kiln water hazard accident will occur based on this. Among them, the parameters in the old kiln water hazard prediction model are all optimized by the genetic algorithm GA, effectively improving the prediction accuracy of the old kiln water hazard prediction model. Through this method, it is possible to achieve high-precision prediction of old kiln water hazard accidents in the area to be predicted, thereby improving the safety level, effectively reducing the occurrence of old kiln water hazard accidents, and ensuring coal mine safety. Description of the Drawings

[0057] The drawings here are incorporated into the specification and form a part of this specification, showing the embodiments that conform to the present application, and are used together with the specification to explain the principles of the present application.

[0058] Figure 1 It is a schematic diagram of the application scenario of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by the present application;

[0059] Figure 2 It is a schematic flow chart of the first embodiment of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by the present application;

[0060] Figure 3 It is a schematic flow chart of the second embodiment of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by the present application;

[0061] Figure 4 It is a schematic flow chart of the third embodiment of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by the present application;

[0062] Figure 5 It is a schematic flow chart of the fourth embodiment of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by the present application;

[0063] Figure 6 Schematic flowchart of Embodiment 5 of an intelligent coal mine water hazard prediction method based on artificial intelligence provided by this application;

[0064] Figure 7 Schematic flowchart of Embodiment 6 of an intelligent coal mine water hazard prediction method based on artificial intelligence provided by this application;

[0065] Figure 8 Schematic flowchart of Embodiment 7 of an intelligent coal mine water hazard prediction method based on artificial intelligence provided by this application;

[0066] Figure 9 Schematic diagram of the framework of an intelligent coal mine water hazard early warning system based on artificial intelligence provided by this application;

[0067] Figure 10 Schematic flowchart of an intelligent coal mine water hazard prediction method based on artificial intelligence provided by this application;

[0068] Figure 11 Schematic diagram of the structure of Embodiment 1 of an intelligent coal mine water hazard prediction device based on artificial intelligence provided by this application;

[0069] Figure 12 Schematic diagram of the structure of Embodiment 2 of an intelligent coal mine water hazard prediction device based on artificial intelligence provided by this application;

[0070] Figure 13 Schematic diagram of the structure of a computer device provided by this application;

[0071] Figure 14 An early warning system for coal mine water hazards based on artificial intelligence provided by this application.

[0072] Through the above-mentioned drawings, the specific embodiments of this application have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of this application in any way, but to illustrate the concept of this application to those skilled in the art by referring to specific embodiments. Detailed implementation manners

[0073] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all the implementation manners consistent with this application. On the contrary, they are merely examples of devices and methods consistent with some aspects of this application as detailed in the appended claims.

[0074] First, the terms involved in this application are explained:

[0075] Artificial Intelligence: It refers to a branch of computer science that aims to research and develop methods, technologies, and application systems capable of simulating, extending, and expanding human intelligence;

[0076] Genetic Algorithm (GA for short): It refers to a computational model that simulates the biological evolution process of Darwin's theory of biological evolution and genetic mechanisms. As a bionic global optimization algorithm, its basic principle is to encode a set of problems into an initial population of chromosomes, calculate the fitness of each individual in the population, and perform operations such as selection, crossover, and mutation on the individuals in the population based on fitness, so that the fitness of the individuals continuously improves until the pre-set termination condition is reached. The algorithm finds the optimal chromosome and decodes it to obtain the optimal solution to the problem;

[0077] Long Short-Term Memory (LSTM for short): It refers to a variant form of the recurrent neural network. By introducing a gating mechanism in the structure of neurons, it selectively decides which data information to add and selectively decides to discard the data information accumulated before, which well solves the long-term dependence problem of the recurrent neural network;

[0078] Selection operation: It refers to the operation in GA that "survives the fittest and eliminates the unfit" for the population, that is, selects individuals with higher fitness and inherits part of their own information to the next generation population. This genetic operation relies on a specific method, which is the selection operation;

[0079] Roulette wheel method: It means that the probability of a certain individual being selected is determined by the ratio of the fitness value of this individual to the fitness value of the entire population. That is, in the same population, the larger the fitness value of this individual, the easier it is to be selected as a parent to inherit its own characteristics to the offspring;

[0080] Crossover operation: It refers to gene recombination. After individuals are binary encoded into chromosomes and then through the selection operation, a part of the parent individuals that can be used for mating are selected. The crossover operation refers to selecting two of these parent individuals to mate with each other, that is, exchanging part of their genes in a certain way according to their chromosomes to form new individuals;

[0081] Mutation operation: The mutation operation is similar to the crossover operation. The difference is that the crossover operation is an operation between two individuals, while the mutation operation is an operation on a single individual itself. Commonly used mutation operations include real number mutation method and binary mutation method, etc.;

[0082] Old kiln water: Also known as old goaf water, it refers to the accumulated water formed when some goaf areas formed during past coal mining operations are filled with groundwater or surface water. This accumulated water mainly exists in the shallow parts of many old mining areas, including flooded mine shafts and roadways, forming a complex interconnected network. Old kiln water is characterized by large water volume, strong acidity, and high water pressure. If the mine shafts and roadways under mining approach or pass through old goaf areas, there is a possibility of water inrush accidents. In addition, the acidic old kiln water directly harms the surrounding environment and may lead to sudden water disaster accidents;

[0083] Four followings: It refers to following measurement, following reporting, following sorting, and following analysis.

[0084] Artificial intelligence technology plays an important role in the intelligent development of coal mines. In mine applications, artificial intelligence technology usually shows the characteristics of point - by - point combination and shallow combination, and has not yet achieved deep integration with the mine production and management system levels, resulting in frequent occurrence of coal mine water disaster accidents.

[0085] To reduce the occurrence of coal mine water disaster accidents, coal mine water disaster detection technologies have gradually developed. Among them, seismic exploration technology and electromagnetic exploration technology are the main detection methods. Seismic exploration technology includes high - precision three - dimensional seismic exploration technology and underground trough - wave seismic detection technology. Electromagnetic exploration technology includes transient electromagnetic detection, high - density resistivity method detection, large - penetration - distance multi - frequency synchronous radio wave perspective and other technologies. Before the application of these technologies, neural network technologies such as backpropagation algorithm and pruning Bayesian neural network are introduced. Neural network technology is based on neural network models and trained with a large amount of coal mine water disaster data to obtain models for predicting coal mine water disaster data. The wide application of these technologies can help coal mine enterprises timely master the situation of coal mine water disasters and take corresponding measures to avoid accidents.

[0086] However, the above - mentioned coal mine water disaster detection technologies still have errors in predicting coal mine water disaster data and fail to effectively avoid the occurrence of coal mine water disaster accidents.

[0087] In view of the above problems, during the research on the prediction of coal mine water disaster data, the inventor found that the current coal mine water disaster detection technology has a strong dependence on the neural network model, and the model parameters are usually determined based on manual experience, which limits the accuracy and stability of the coal mine water disaster detection technology. Therefore, the inventor considered whether it is possible to improve the prediction accuracy of coal mine water disaster data from the perspectives of the selection of the neural network model and the determination of parameters. Specifically, the perception layer device collects the old goaf water monitoring data in the first preset time period before the current moment in the area to be monitored in the coal mine in real time, and transmits it to the old goaf water disaster prediction model in the computer device for processing, so as to obtain the predicted data of the water inflow of the working face in the second preset time period after the current moment. Among them, the old goaf water disaster prediction model uses the LSTM model as the initial model structure and optimizes the initial model parameters with GA to improve the prediction accuracy of the old goaf water disaster prediction model, thereby reducing the probability of old goaf water disaster accidents to a certain extent and ensuring the safety of coal mines.

[0088] Figure 1 The figure is a schematic diagram of the application scenario of an intelligent prediction method for coal mine water disasters based on artificial intelligence provided by this application. As Figure 1 shown, the application scenario of the solution provided by this application includes a perception layer device 101 and a computer device 102. Among them, the perception layer device 101 is mainly used to collect the old goaf water monitoring data in the area to be predicted in the coal mine, and the old goaf water disaster prediction model 1021 pre-deployed in the computer device 102 is mainly used to obtain the old goaf water monitoring data, process the old goaf water monitoring data, and judge whether an old goaf water disaster accident may occur.

[0089] The perception layer device 101 includes, but is not limited to, temperature sensors, pressure sensors, water level sensors, pH sensors, water conduction speed sensors, water hardness sensors, dissolved oxygen sensors, ammonia nitrogen compound concentration sensors, turbidity sensors, water inflow sensors of the working face, water quality sensors, and old goaf damage depth sensors, etc. The computer device 102 includes, but is not limited to, servers and personal computers, etc. Although Figure 1 only one perception layer device 101 and one computer device 102 are shown in the figure, it should be understood that there may be two or more perception layer devices 101 and computer devices 102.

[0090] Specifically, in the area to be predicted in a coal mine, the sensing layer device 101 collects the old kiln water monitoring data within the first preset time period before the current moment in real time, and transmits it to the computer device 102 through the gateway system. After being processed by the old kiln water hazard prediction model 1021 deployed in the computer device 102, the prediction of the water inflow data of the working face within the second preset time period after the current moment in the area to be predicted is realized. This prediction process provides data support for judging whether an old kiln water hazard accident may occur, so as to take corresponding preventive measures when necessary and reduce the probability of old kiln water hazard accidents.

[0091] The following uses specific embodiments to elaborate in detail on the technical solution of the present application and how the technical solution of the present application solves the above technical problems. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0092] Figure 2 It is a schematic flowchart of the first embodiment of an intelligent coal mine water hazard prediction method based on artificial intelligence provided by the present application. As Figure 2 shown, the process of the intelligent coal mine water hazard prediction method based on artificial intelligence may include:

[0093] S201: Receive the old kiln water monitoring data in the area to be predicted in the coal mine within the first preset time period before the current moment uploaded by the sensing layer device, where the old kiln water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old kiln damage monitoring data.

[0094] In this step, in order to reduce the occurrence of old kiln water hazard accidents in the area to be predicted in the coal mine, the real-time acquisition and processing of old kiln water monitoring data are particularly crucial. Among them, the area to be predicted in the coal mine can be a single area or multiple areas. The area to be predicted in the coal mine refers to the old working area or the goaf area of the coal mine, where the accumulation of old kiln water may exist.

[0095] Specifically, the old kiln water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old kiln damage monitoring data.

[0096] Optionally, the water source monitoring data mainly includes the monitoring indicators of the water source, including but not limited to temperature, pressure, water level, acidity and alkalinity, conductivity, total hardness, dissolved oxygen, ammonia nitrogen compounds, and turbidity, etc. By monitoring these indicators, it can help to understand the state and change trend of the coal mine groundwater system.

[0097] The monitoring data of water inrush in the working face mainly includes the monitoring indicators of water inrush in the working face, including but not limited to the water inrush volume, water quality, temperature, etc. By monitoring these indicators, water inrush problems can be detected in a timely manner, so as to take corresponding measures for treatment and ensure the safe production of the mine.

[0098] The monitoring data of old kiln damage mainly includes the monitoring indicators of old kiln damage, including but not limited to the depth of old kiln damage, etc. By monitoring these indicators, potential hidden dangers of geological disasters can be detected in a timely manner, which is conducive to taking measures for treatment and prevention and avoiding the occurrence of coal mine safety accidents.

[0099] In order to obtain the old kiln water monitoring data in the first preset time period before the current moment in the area to be predicted in the coal mine in a timely manner, at least one sensing layer device is pre-installed in the area to be predicted in the coal mine. Among them, the sensing layer devices include but not limited to temperature sensors, pressure sensors, water level sensors, pH sensors, water conduction speed sensors, water hardness sensors, dissolved oxygen sensors, ammonia nitrogen compound concentration sensors, turbidity sensors, water inrush volume sensors in the working face, water quality sensors, and old kiln damage depth sensors, etc.

[0100] After the sensing layer device obtains the old kiln water monitoring data in the first preset time period before the current moment in the area to be predicted in the coal mine, it will transmit the obtained old kiln water monitoring data to the computer device through the gateway system in a timely manner for prediction and analysis of the old kiln water monitoring data.

[0101] S202: Input the old kiln water monitoring data in the first preset time period in the area to be predicted into the old kiln water hazard prediction model for prediction processing to obtain the predicted data of the water inrush volume in the working face in the second preset time period after the current moment. Among them, the old kiln water hazard prediction model is an LSTM model for predicting the water inrush volume in the working face that is pre-optimized based on GA, and GA is used to determine the parameters of the LSTM model; the length of the first preset time period is greater than the length of the second preset time period.

[0102] In this step, based on step S201, after the computer device receives the old kiln water monitoring data in the first preset time period before the current moment in the area to be predicted in the coal mine uploaded by the sensing layer device, it will input the old kiln water monitoring data into the old kiln water hazard prediction model for prediction processing in order to obtain the predicted data of the water inrush volume in the working face in the second preset time period after the current moment.

[0103] Specifically, the old kiln water hazard prediction model is an LSTM model optimized in advance based on GA for predicting the water inflow of the working face. By analyzing the trends and patterns of historical old kiln water monitoring data and combining the old kiln water monitoring data within the first preset time period uploaded by the devices in the perception layer at the current moment, this model can accurately predict the predicted water inflow data of the working face in the second preset time period after the current moment in the area to be predicted in the coal mine.

[0104] Among them, the LSTM model is suitable for processing and predicting time series data, has strong memory, and can capture long-term dependence relationships in time series. Before applying the LSTM model to predict the predicted water inflow data of the working face, it is necessary to determine the parameters in the LSTM model in advance. These parameters include but are not limited to hyperparameters, weight parameters, and bias term parameters in the LSTM model, etc.

[0105] In the existing technical solutions, the hyperparameters in the LSTM model are determined by artificial experience, and parameters such as weights and bias terms are obtained by training the LSTM model with a large amount of historical data. Among them, the method of determining hyperparameters by artificial experience usually lacks consistency and accuracy, and there is usually a problem of high time complexity in obtaining parameters such as weights and bias terms by training the LSTM model with historical data.

[0106] Therefore, this application proposes to determine the parameters in the LSTM model by means of GA. Among them, GA is an optimization algorithm that simulates natural selection and genetic mechanisms, and determines the parameters in the LSTM model by simulating the process of evolution, so as to improve the prediction accuracy and reliability of the old kiln water hazard prediction model.

[0107] It should be noted that the length of the first preset time period needs to be greater than the length of the second preset time period. Specifically, the monitoring data within the longer first preset time period can provide historical old kiln water monitoring data within a longer time range, which is used to establish and verify the old kiln water hazard prediction model and judge the accuracy and reliability of the prediction results. The length of the second preset time period is relatively short, which is to ensure that the management personnel can timely understand whether there will be old kiln water hazard accidents in the area to be predicted in the current coal mine, so that the management personnel can take corresponding measures for prevention and treatment.

[0108] By using the monitoring data of the longer first preset time period, the model can be trained and optimized more fully to improve the accuracy of the prediction results. This can help the management personnel better understand the development trend and possible impacts of the old kiln water hazard, so as to take measures in time to reduce the probability of accidents. The shorter second preset time period can provide real-time prediction results, enabling the management personnel to make decisions and take actions quickly to deal with emergencies and ensure the safety of the working face.

[0109] In summary, the length of the first preset time period is greater than that of the second preset time period, which can ensure that there is sufficient historical data support during the establishment and optimization of the prediction model, and can timely give early warnings and take measures to deal with the predicted results.

[0110] S203: Determine whether an old kiln water hazard accident will occur according to the predicted data of the water inflow of the working face in the second preset time period in the area to be predicted.

[0111] In this step, according to step S202, when the computer device obtains the predicted data of the water inflow of the working face in the second preset time period in the area to be predicted, it determines whether there is a potential old kiln water hazard accident in the area to be predicted according to the predicted data of the water inflow of the working face.

[0112] Specifically, by comparing the predicted data of the water inflow of the working face with the preset threshold value of the water inflow data of the working face, it can be accurately determined which areas may have old kiln water hazard accidents, so as to take corresponding prevention and control measures to ensure safety and controllability.

[0113] The intelligent prediction method for coal mine water hazards based on artificial intelligence provided in the embodiments of the present application takes the old kiln water hazard prediction model as the core, processes the old kiln water monitoring data in the first preset time period before the current moment obtained by the perception layer device in real time, and obtains the predicted data of the water inflow of the working face in the second preset time period after the current moment in the area to be predicted, so as to accurately judge whether an old kiln water hazard accident will occur. Through this solution, the old kiln water monitoring data can be effectively utilized, potential old kiln water hazard accidents can be discovered in time, and thus the possibility of old kiln water hazard accidents occurring can be minimized to the greatest extent, and the safety level of coal mines can be improved.

[0114] Figure 3 This is a schematic flowchart of the second embodiment of the intelligent prediction method for coal mine water hazards based on artificial intelligence provided by the present application. As Figure 3 shown, on the basis of the above embodiments, the process of the intelligent prediction method for coal mine water hazards based on artificial intelligence may include:

[0115] S301: Determine the target parameter configuration scheme of the LSTM model according to GA.

[0116] In this step, the determination of the old kiln water hazard prediction model is mainly based on GA.

[0117] Specifically, GA is used to optimize and select the target parameters in the LSTM model, so as to determine the target parameter configuration scheme of the LSTM model.

[0118] Optionally, the target parameters include, but are not limited to, the number of hidden layers in the LSTM model, the number of neurons in each hidden layer, the learning rate, the weight parameters, and the regularization parameters, etc. These target parameters constitute the target parameter configuration scheme.

[0119] Among them, before determining the target parameter configuration scheme of the LSTM model, it is necessary to build the neural network structure of the LSTM model. Specifically, the neural network structure of the LSTM model includes an input gate, a forget gate, a candidate memory cell, a memory cell, an output gate, and a hidden state. These parts work together to enable the LSTM to effectively capture long-term dependencies, thereby efficiently processing the old goaf water monitoring data.

[0120] S302: Configure the target parameters in the LSTM model according to the target parameter configuration scheme to obtain the old goaf water hazard prediction model.

[0121] In this step, according to step S301, after determining the target parameter configuration scheme of the LSTM model, the target parameters in the LSTM model are configured according to the target parameter configuration scheme, thereby obtaining the old goaf water hazard prediction model.

[0122] Specifically, according to the predefined neural network structure and target parameter configuration scheme of the LSTM model, the parameters at the corresponding positions in the LSTM model are determined, thereby obtaining the old goaf water hazard prediction model for predicting the water inflow data of the working face.

[0123] The intelligent coal mine water hazard prediction method based on artificial intelligence provided in the embodiments of the present application mainly provides the acquisition process of the old goaf water hazard prediction model. Among them, the basic framework of the old goaf water hazard prediction model is the LSTM model, and the target parameters in the LSTM model are determined by GA optimization. Through this method, the target parameters in the old goaf water hazard prediction model can be accurately determined, and the prediction accuracy and reliability of the old goaf water hazard prediction model can be improved.

[0124] Figure 4 This is a schematic flowchart of Embodiment 3 of the intelligent coal mine water hazard prediction method based on artificial intelligence provided by the present application. As Figure 4 shown, on the basis of the above embodiments, according to GA, determine the target parameter configuration scheme of the LSTM model. The process of this intelligent coal mine water hazard prediction method based on artificial intelligence can include:

[0125] Step a: Obtain a sample data set based on historical old kiln water monitoring data. The historical old kiln water monitoring data includes all historical water source monitoring data, historical working face water inrush monitoring data, and historical old kiln water damage monitoring data. The sample data set includes multiple groups of data samples. Each group of data samples includes old kiln water monitoring data for two consecutive time periods, and the duration of the first time period is greater than or equal to the duration of the first preset time period, and the duration of the second time period is the same as the duration of the second preset time period.

[0126] In this step, if you want to determine the target parameter configuration scheme of the LSTM model, you need to obtain historical old kiln water monitoring data in advance.

[0127] Among them, the historical old kiln water monitoring data includes all historical water source monitoring data, historical working face water inrush monitoring data, and historical old kiln water damage monitoring data. After obtaining the historical old kiln water monitoring data, a sample data set is determined based on the obtained historical old kiln water monitoring data.

[0128] In the sample data set, there are multiple groups of data samples. Each group of data samples includes old kiln water monitoring data for two consecutive time periods. Among them, in order to accurately predict the predicted water inrush volume data in the second preset time period, the duration of the first time period is preset to be greater than or equal to the duration of the first preset time period, and the duration of the second time period is the same as the duration of the second preset time period.

[0129] Optionally, before obtaining the sample data set based on the historical old kiln water monitoring data, preprocessing operations can be performed on the historical old kiln water monitoring data. For example, filling missing data and data standardization, etc. Specifically, due to certain reasons, there are missing values in the historical old kiln water monitoring data. To ensure the scientific nature and effectiveness of data processing and thus obtain reasonable judgment results, scientific methods need to be used for processing. In order to improve the training effect (unify the dimension) and efficiency (quickly converge) of the old kiln water hazard prediction model, the z-score standardization method is used to standardize each feature column of the historical old kiln water monitoring data.

[0130] Step b: Initialize N groups of first parameter configuration schemes of the GA, selection operator, adaptive crossover probability, adaptive mutation probability, and preset maximum number of iterations, where N is a positive integer. Each group of first parameter configuration schemes includes the number of hidden layer neurons, learning rate, preset number of iterations, minimum batch size, regularization parameter, weight parameter, and bias term parameter.

[0131] In this step, before applying the GA to determine the target parameter configuration scheme of the LSTM model, it is necessary to pre-determine the parameters of the GA. Among them, the parameters of the GA include but are not limited to N groups of first parameter configuration schemes, selection operator, adaptive crossover probability, adaptive mutation probability, and preset maximum number of iterations, etc.

[0132] Specifically, each group of the first parameter configuration schemes includes the hyperparameters and parameters of the LSTM. Among them, the hyperparameters include the number of neurons in the hidden layer, learning rate, preset number of iterations, mini-batch size, and regularization parameter. The parameters include weight parameters and bias term parameters.

[0133] The number of neurons in the hidden layer refers to the number of neurons in the hidden layer of the neural network. The role of the hidden layer is to perform non-linear transformation on the input data to learn the complex features of the data.

[0134] The number of neurons in the hidden layer will affect the capacity and complexity of the LSTM model. The learning rate refers to the amplitude that controls the parameter update in each iteration in the gradient descent algorithm, which mainly affects the training effect and speed of the LSTM model.

[0135] The preset number of iterations refers to the number of iterations when training the LSTM model. During the training process, the LSTM model will repeatedly use the training data for parameter update, and the preset number of iterations determines how many times the LSTM model will perform parameter update operations.

[0136] The mini-batch size refers to the number of samples used to update the parameters of the LSTM model at one time when training the LSTM model. The regularization parameter is used to control the complexity of the LSTM model and prevent overfitting parameters.

[0137] Regularization limits the size of the LSTM model parameters by adding a penalty term to the loss function, avoiding overfitting of the LSTM model on the training data, and improving the generalization ability of the LSTM model.

[0138] In the LSTM model, the weight parameters include the weights of the input gate, forget gate, output gate, and cell state, which are used to control the influence of input, forgetting, and output on the cell state.

[0139] The bias term parameters include the bias terms of the input gate, forget gate, output gate, and cell state, which are used to adjust the thresholds for the corresponding gate control and cell state update.

[0140] Optionally, the value range of N is generally 20 to 100. If the value is small, it is beneficial to improve the operation and convergence speed of the GA, but the diversity of the population may not be satisfied. If the value is large, it is beneficial to improve the search quality of the GA, but it will increase the computational amount and lead to a decrease in the convergence speed. Generally, the value is 64.

[0141] Step c: Construct n LSTM models with exactly the same structure to obtain n first LSTM models.

[0142] In this step, in order to determine the target parameter configuration scheme of the LSTM model, LSTM models with the same number as the number of the first parameter configuration schemes are pre-constructed to obtain N first LSTM models.

[0143] Among them, the structures of the N LSTM models are exactly the same, laying a foundation for the target parameter configuration scheme of the LSTM model.

[0144] Step d: Perform chromosome encoding on each group of the first parameter configuration schemes to obtain corresponding first chromosomes.

[0145] In this step, according to step b, after initializing N groups of the first parameter configuration schemes of the GA, chromosome encoding is performed on each group of the first parameter configuration schemes, thereby obtaining N first chromosomes.

[0146] Specifically, for each group of the first parameter configuration schemes, the parameter values therein can be encoded according to certain rules, such as using binary, integer encoding, or real number encoding, etc., to obtain corresponding chromosome representations. The purpose of doing this is to enable the GA to optimize and analyze these parameters, and through the evolutionary process of the population, find the target parameter configuration scheme.

[0147] Step e: For each first LSTM model, select a first chromosome corresponding to the first parameter configuration scheme, and configure the initial parameters in the first LSTM model to obtain n second LSTM models. Among them, the initial parameters of the first LSTM model include the number of initial hidden layer neurons, the initial learning rate, the initial preset number of iterations, the initial mini-batch size, the initial regularization parameter, the initial weight parameter, and the initial bias term parameter.

[0148] In this step, after obtaining N first LSTM models based on step c, for each first LSTM model, select a first chromosome corresponding to the first parameter configuration scheme, and configure the initial parameters in the first LSTM model, thereby obtaining N second LSTM models.

[0149] Among them, the initial parameters in the first LSTM model include the number of initial hidden layer neurons, the initial learning rate, the initial preset number of iterations, the initial mini-batch size, the initial regularization parameter, the initial weight parameter, and the initial bias term parameter. These initial parameters are consistent with the parameters in the first parameter configuration scheme corresponding to the first chromosome in terms of quantity and meaning.

[0150] Step f: Select N groups of sample data from the sample data set, and input the historical old kiln water monitoring data in the first time period of the N groups of sample data into N second LSTM models respectively for prediction, so as to obtain N predicted data of the water inflow of the working face in the second time period.

[0151] In this step, after obtaining N second LSTM models according to step e, select N groups of sample data from the sample data set obtained in step a, and input the historical old kiln water monitoring data in the first time period of the N groups of sample data into N second LSTM models respectively for prediction, so as to obtain N predicted data of the water inflow of the working face in the second time period. In this step, after obtaining N second LSTM models according to step e, these second LSTM models need to be applied to the actual data set.

[0152] Specifically, select N groups of sample data from the sample data set obtained in step a. Each group of sample data includes historical old kiln water monitoring data and corresponding water inflow data of the working face. For each group of sample data, input the historical old kiln water monitoring data in the first time period into the corresponding second LSTM model for prediction, that is, apply each second LSTM model to a specific sample data set to generate the predicted data of the water inflow of the working face in the corresponding time period.

[0153] Through this method, N predicted data of the water inflow of the working face in the second time period can be obtained. These prediction results reflect the prediction ability of each second LSTM model for the given input data.

[0154] Step g: Calculate the fitness value of each first chromosome according to the predicted data of the water inflow of the working face in the N second time periods and the preset fitness function of GA.

[0155] In this step, calculate the fitness value of each first chromosome according to the N predicted data of the water inflow of the working face in the second time period obtained in step f and the preset fitness function of GA.

[0156] In a specific implementation manner, the fitness value of each first chromosome is calculated by the following formula:

[0157]

[0158] where represents the fitness value of the nth first chromosome, and f MAE (n) represents the mean absolute error value of the nth first chromosome, and f RMSE (n) represents the root mean square error value of the nth first chromosome.

[0159] Specifically, f MAE(n) is calculated as follows:

[0160]

[0161] Where N represents the number of the first chromosomes, represents the true data of the water inflow of the working face in the second time period corresponding to the nth first chromosome, represents the predicted data of the water inflow of the working face in the second time period corresponding to the nth first chromosome.

[0162] f RMSE (n) is calculated as follows:

[0163]

[0164] The calculation method of the fitness value fully considers the gap between the predicted data of the water inflow of the working face in the second time period and the true data of the water inflow of the working face in the second time period. Through this method, the performance of each group of the first parameter configuration schemes can be comprehensively judged, so as to better optimize the search process of the GA to determine the target parameter configuration scheme.

[0165] For the intelligent coal mine water disaster prediction method based on artificial intelligence provided in this embodiment, according to the mean absolute error value and root mean square error value of each first chromosome, the fitness of each first chromosome can be calculated. The size of the fitness directly reflects the quality of the first parameter configuration scheme represented by the first chromosome. By calculating the fitness, the advantages and disadvantages of different parameter configuration schemes can be quantified and compared, so as to provide an objective basis for selecting the target parameter configuration scheme.

[0166] Step h: According to the fitness value of each first chromosome, sort the fitness values of the N first chromosomes from high to low, obtain the number of the first chromosome corresponding to the highest fitness value, and record it as the target chromosome number.

[0167] In this step, after obtaining the fitness value of each first chromosome according to step g, then according to the fitness value of each first chromosome, sort the fitness values of the N first chromosomes, arrange them from high to low. By doing so, the first chromosome with the highest fitness value can be found, and then the target chromosome number can be determined.

[0168] Where the target chromosome number corresponds to the first chromosome with the highest fitness value. By recording the target chromosome number, it can be clearly understood which first parameter configuration scheme is the most excellent at the current stage.

[0169] Step i: According to the fitness value corresponding to each first chromosome and the selection operator, perform screening processing on the first chromosomes to obtain N screened first chromosomes.

[0170] In this step, after obtaining the fitness value of each first chromosome according to step g, the first chromosomes are screened according to the fitness value corresponding to each first chromosome and the selection operator, so as to obtain N screened first chromosomes.

[0171] Figure 5 It is a schematic flow chart of the fourth embodiment of an artificial intelligence-based intelligent prediction method for coal mine water hazards provided by this application. In a specific implementation manner, the specific process of screening the first chromosomes according to the fitness value corresponding to each first chromosome and the selection operator is as Figure 5 shown. The flow of this artificial intelligence-based intelligent prediction method for coal mine water hazards may include:

[0172] S501: Among all the first chromosomes, the first chromosomes whose fitness values meet the preset fitness condition are determined as the first part of chromosomes.

[0173] In this step, after calculating the fitness values corresponding to all the first chromosomes, among all the first chromosomes, the first chromosomes whose fitness values meet the preset fitness condition are determined as the first part of chromosomes.

[0174] In a possible implementation manner, the preset fitness condition may be to sort all the first chromosomes in descending order of fitness value, and the first preset number of first chromosomes are used as the first part of chromosomes, or the first chromosomes whose fitness values are greater than the preset fitness value are used as the first part of chromosomes.

[0175] S502: According to the roulette wheel method, the second part of chromosomes is determined from all the first chromosomes. The N screened first chromosomes include the first part of chromosomes and the second part of chromosomes.

[0176] In this step, according to step S501, after determining the first chromosomes that meet the preset fitness condition as the first part of chromosomes, the remaining chromosomes are used to determine the second part of chromosomes from all the first chromosomes by means of the roulette wheel method.

[0177] Among them, the roulette wheel method is a selection method based on fitness values, which simulates the operation principle of a roulette wheel, so that the first chromosomes with higher fitness values have a greater probability of being selected. The specific steps are as follows: First, calculate the sum of the fitness values of all the first chromosomes. Then, according to the proportion of the fitness value of each first chromosome in the sum, construct the corresponding interval range, and this interval represents the probability of each first chromosome being selected. Finally, a random number between 0 and 1 is generated as the roulette pointer, and the selected first chromosome is determined according to the interval where the pointer falls. It should be noted that the interval range corresponding to the first chromosomes with higher fitness values is larger, so their probability of being selected is also higher.

[0178] Optionally, both the first part of the chromosome and the second part of the chromosome account for a preset proportion of the total number of the first chromosome. Generally, the number of the second part of the chromosome determined by the preset fitness condition accounts for 20% of the total number of the first chromosome, and the number of the second part of the chromosome determined by the roulette method accounts for 80% of the total number of the first chromosome. The specific preset proportion can be determined according to the actual situation and is not limited herein.

[0179] It should be noted that, in addition to the roulette method, the determination method of the second part of the chromosome can also be the ranking selection method, the random selection method, etc. Among them, the ranking selection method is a ranking-based selection method, which mainly sorts all chromosomes according to the fitness value and determines the probability of each chromosome being selected according to the level of fitness. The random selection method means randomly selecting a certain number of chromosomes from the current population as the second part of the chromosome.

[0180] After the first part of the chromosome and the second part of the chromosome are determined, the N screened first chromosomes include the first part of the chromosome and the second part of the chromosome.

[0181] The intelligent coal mine water disaster prediction method based on artificial intelligence provided in this embodiment mainly provides the selection operation of GA. Among them, after calculating the fitness values of all the first chromosomes, the first part of the chromosome is determined according to the preset adaptation condition, and the second part of the chromosome is determined according to the roulette method, so as to determine the N screened first chromosomes. Through this method, it can be ensured that excellent chromosomes will not be eliminated during the genetic process, effectively maintaining the high fitness level of the population, and then improving the convergence speed and global search ability of the genetic algorithm, so as to find the target parameter configuration scheme faster.

[0182] Step j: Calculate the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome according to the fitness value of each screened first chromosome.

[0183] In this step, after obtaining the N screened first chromosomes according to step i, the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome are calculated according to the fitness value of each screened first chromosome.

[0184] Among them, in each iteration process, the selection of the crossover probability and the mutation probability is the key to affecting the behavior and performance of GA, which will directly affect the convergence of GA.

[0185] This application sets the adaptive crossover probability and the adaptive mutation probability, which can ensure that GA jumps out of the local optimum situation and at the same time ensure the survival of excellent chromosomes. The specific implementation method can refer to step S601 to step S602 and will not be elaborated herein.

[0186] Step k: Based on the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome, perform adaptive crossover and mutation operations on the screened first chromosomes to generate N second chromosomes.

[0187] In this step, after calculating the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome according to Step j, perform adaptive crossover and mutation operations on each screened first chromosome to generate N second chromosomes.

[0188] Specifically, first, according to the adaptive crossover probability of each first chromosome, select two chromosomes from the N screened first chromosomes for crossover operation. These two selected first chromosomes will serve as the parent chromosomes. Among them, the first chromosome with a high fitness value will have a higher probability of being selected as the parent chromosome to ensure the inheritance of excellent genes. Then, through the chromosome crossover method, exchange part of the gene fragments of the two parent chromosomes to generate a new second chromosome, which will serve as the offspring chromosome. Finally, according to the adaptive mutation probability of each chromosome, perform mutation operations on each second chromosome.

[0189] In a possible implementation manner, the crossover operation is implemented by using the single-point crossover method. Specifically, only randomly set a crossover point in the coding string corresponding to each first chromosome, divide the first chromosome into two parts, and the left and right sides of the offspring chromosome come from the two parent chromosomes respectively.

[0190] In another possible implementation manner, the binary mutation method is applied to implement the mutation operation. Specifically, perform the inverse operation of changing "0" to "1" or changing "1" to "0" on the selected gene.

[0191] Through the crossover and mutation operations, a comprehensive exploration of the search space is achieved, which helps the GA find the target parameter configuration scheme faster.

[0192] Optionally, before performing the crossover operation, the number of adaptive crossover points and the position of the crossover points are predetermined. Specifically, after calculating the adaptive crossover probability of each first chromosome after screening, a random number method is first used to generate an adaptive crossover point number, wherein the adaptive crossover point number is greater than or equal to the preset minimum crossover point number and less than or equal to the preset maximum crossover point number. Then, based on the generated adaptive crossover point number and the chaotic sequence generation algorithm, a random number sequence uniformly distributed in the range of [0, 1] is obtained, wherein the number in the random number sequence is consistent with the number of adaptive crossover points. Each value in the random number sequence is then multiplied and rounded by the length of each first chromosome after screening to obtain the position of the crossover point. Finally, based on the position of the crossover point, a crossover operation is performed to generate a new second chromosome.

[0193] Optionally, before performing the mutation operation, the number and location of adaptive mutation points are pre-determined. Specifically, after calculating the adaptive mutation probability of each screened first chromosome, a random number method is first used to generate an adaptive number of mutation points, wherein the adaptive number of mutation points is greater than or equal to a preset minimum number of mutation points and less than or equal to a preset maximum number of mutation points. Then, based on the generated number of adaptive mutation points and a chaotic sequence generation algorithm, a random number sequence uniformly distributed within the range [0, 1] is obtained, wherein the number in the random number sequence is consistent with the number of adaptive mutation points. Each value in the random number sequence is then multiplied and rounded by the length of each screened first chromosome to obtain the location of the mutation point. Finally, a mutation operation is performed based on the location of the mutation point to generate a new second chromosome. Step 1: Determine whether the current number of iterations has reached a preset maximum number of iterations. If not, the N second chromosomes are determined as the new n first chromosomes. Steps e through k are repeated until the preset maximum number of iterations is reached. When the preset maximum number of iterations is reached, the first parameter configuration scheme corresponding to the target chromosome number is determined as the target parameter configuration scheme.

[0194] In this step, after obtaining N second chromosomes according to step k, it is determined whether the current number of iterations reaches the preset maximum number of iterations.

[0195] If the preset maximum number of iterations is not reached, the N second chromosomes are determined as new N first chromosomes, and steps e to k are repeated until the preset maximum number of iterations is reached.

[0196] If the preset maximum number of iterations is reached, the first parameter configuration scheme corresponding to the target chromosome is determined as the target parameter configuration scheme. The target parameter configuration scheme is a better solution after the iteration and optimization process and can be used to configure the target parameters in the LSTM model.

[0197] For example, assume that the preset maximum number of iterations is 500 times. If the current number of iterations is 345 times, then steps e to k are continued. If the current number of iterations is 500 times, then the first parameter configuration scheme corresponding to the obtained target chromosome is determined as the target parameter configuration scheme.

[0198] The intelligent coal mine water hazard prediction method based on artificial intelligence provided by the embodiments of the present application uses GA to determine the target parameter configuration scheme of the LSTM model. Specifically, according to the preset maximum number of iterations, the operations of chromosome selection, crossover, and mutation are repeatedly executed until the target parameter configuration scheme is found. Through this iterative optimization method, GA can gradually converge to the optimal solution during continuous iteration, thus laying a foundation for obtaining the old kiln water hazard prediction model.

[0199] Next, the specific implementation process of step j will be elaborated. Figure 6 It is a schematic flowchart of Embodiment 5 of the intelligent coal mine water hazard prediction method based on artificial intelligence provided by the present application. As Figure 6 shown, on the basis of the above embodiments, according to the fitness value of each filtered first chromosome, the adaptive crossover probability and the adaptive mutation probability of each filtered first chromosome are calculated. The process of the intelligent coal mine water hazard prediction method based on artificial intelligence may include:

[0200] S601: For each filtered first chromosome, according to the preset initial crossover probability, the current number of iterations, the preset maximum number of iterations, and the fitness of the filtered first chromosome, determine the adaptive crossover probability of the filtered first chromosome.

[0201] In this step, after GA completes the selection operation to obtain the filtered first chromosome, the crossover operation needs to be further executed.

[0202] Among them, the crossover operation generates new offspring individuals by combining the gene information of two parent individuals. Among them, the crossover probability determines the proportion of individuals in the population that will perform the crossover operation. A higher crossover probability helps to explore a larger solution space, increase the diversity of the population, and thus improve the possibility of finding the global optimal solution. A lower crossover probability helps to maintain the gene structure of the current excellent solution and prevent the problem of the solution quality decline caused by excessive exploration.

[0203] To improve the search ability and convergence speed of GA, prevent premature convergence, enhance robustness, reduce the need for manual parameter adjustment, and improve the quality of the solution, the present application sets the adaptive crossover probability. By dynamically adjusting the crossover probability, it is ensured that GA can adopt an appropriate crossover probability at different stages, so as to more effectively explore the solution space and find the global optimal solution.

[0204] In an alternative embodiment, the calculation formula for the adaptive crossover probability of each filtered first chromosome is as follows:

[0205]

[0206] Wherein, represents the adaptive crossover probability of the filtered first chromosome n, represents the preset initial crossover probability, i represents the current iteration number, i max represents the preset maximum iteration number, fit n represents the fitness value of the filtered first chromosome n, E(fit n ) represents the mathematical expectation of the variable fit n .

[0207] S602: Determine the adaptive mutation probability of the filtered first chromosome according to the preset initial mutation probability, the current iteration number, the preset maximum iteration number, and the fitness of the filtered first chromosome.

[0208] In this step, after the GA completes the crossover operation, a mutation operation needs to be performed.

[0209] Specifically, the mutation operation introduces new gene mutations by randomly changing the gene information of the chromosome. The mutation probability determines what proportion of individuals in the population will undergo the mutation operation. Among them, a higher mutation probability helps to prevent the population from falling into a local optimal solution and increases the diversity of the population, especially in the initial stage of the search. A lower mutation probability helps to achieve stable convergence and avoid excessive random perturbations affecting the quality of the solution, especially in the later stage of the search.

[0210] To improve the performance and effect of the GA, this application sets an adaptive mutation probability to ensure that the GA can adopt an appropriate mutation probability at different stages, thereby effectively exploring the solution space and finding the global optimal solution.

[0211] In an alternative embodiment, the calculation formula for the adaptive mutation probability of each filtered first chromosome is as follows:

[0212]

[0213] Wherein, represents the adaptive mutation probability of the filtered first chromosome n, represents the preset initial mutation probability, i represents the current iteration number, i max represents the preset maximum iteration number, fit n represents the fitness value of the filtered first chromosome n, E(fit n ) represents the mathematical expectation of the variable fit n .

[0214] The intelligent prediction method for coal mine water hazards based on artificial intelligence provided by the embodiments of this application improves the calculation methods of the adaptive crossover probability and the adaptive mutation probability. Specifically, during the execution of GA, based on the fitness of each chromosome, the adaptive crossover probability and the adaptive mutation probability of each chromosome are calculated, effectively overcoming the disadvantages of the traditional GA's poor local optimization ability and easy to fall into local optimum, thus significantly improving the optimization speed of GA and laying a foundation for finding a high-quality target parameter configuration scheme.

[0215] Figure 7 It is a schematic flowchart of Embodiment 6 of the intelligent prediction method for coal mine water hazards based on artificial intelligence provided by this application. As Figure 7 shown, based on any of the above embodiments, the process of the intelligent prediction method for coal mine water hazards based on artificial intelligence further includes:

[0216] S701: If the predicted data of the water inflow of the working face within the second preset time period is lower than the preset water inflow data of the working face, it is determined that no old kiln water hazard accident will occur in the area within the second preset time period.

[0217] In this step, after obtaining the predicted data of the water inflow of the working face within the second preset time after the current moment according to the old kiln water hazard prediction model, the predicted data of the water inflow of the working face and the preset water inflow data of the working face are compared to determine whether an old kiln water hazard accident will occur.

[0218] When the predicted data of the water inflow of the working face within the second preset time period is lower than the preset water inflow data of the working face, it can be determined that no old kiln water hazard accident will occur in the area within the second preset time period.

[0219] Among them, the preset water inflow data of the working face is usually determined according to information such as historical old kiln water monitoring data, geological exploration results, hydrogeological conditions, and relevant engineering experience.

[0220] For example, by analyzing historical old kiln water monitoring data, the typical water inflow range and frequency of the area to be predicted can be understood. Geological exploration work includes geological drilling, core analysis, etc., which can obtain groundwater hydrogeological information and identify possible water sources, water levels, hydrogeological characteristics, etc. Judging the groundwater hydrogeological conditions, including drawing hydrogeological profiles and testing hydrogeological parameters, can understand the groundwater migration law and the recharge source of groundwater, etc. With the help of expert opinions and engineering experience in related fields, the preset water inflow of the working face is determined.

[0221] By comprehensively using the above methods, the preset water inflow data of the working face in the coal mine area to be predicted can be obtained, thus helping to prevent and reduce old kiln water hazard accidents and ensuring the safety of coal mines.

[0222] S702: If the predicted water inflow data of the working face within the second preset time period is higher than or equal to the preset water inflow data of the working face, it is determined that an old kiln water hazard accident will occur in the area within the second preset time period.

[0223] In this step, when comparing the predicted water inflow data of the working face with the preset water inflow data of the working face, if the predicted water inflow data of the working face within the second preset time period is higher than or equal to the preset water inflow data of the working face, it can be determined that an old kiln water hazard accident will occur in the area within the second preset time period.

[0224] Based on this, the management personnel will further take preventive and treatment measures to avoid the occurrence of old kiln water hazard accidents and ensure the safety of the coal mine.

[0225] The intelligent prediction method for coal mine water hazards based on artificial intelligence provided by this application example determines whether an old kiln water hazard accident may occur in the area to be predicted within the second preset time period according to the magnitude relationship between the preset water inflow data of the working face and the real-time obtained water inflow data of the working face. This method can understand the water inflow situation of the working face in the area to be predicted in advance and take preventive and response measures in a timely manner when necessary to ensure the safety of the coal mine.

[0226] Figure 8 This is a schematic flowchart of the seventh embodiment of the intelligent prediction method for coal mine water hazards based on artificial intelligence provided by this application. As Figure 8 shown, on the basis of any of the above embodiments, the process of the intelligent prediction method for coal mine water hazards based on artificial intelligence further includes:

[0227] S801: If it is determined that an old kiln water hazard accident will not occur in the area within the second preset time period, a safety notice message will be pushed to the management personnel, where the safety notice message is used to indicate that an old kiln water hazard accident will not occur in the area.

[0228] In this step, according to step S701, if it is determined that an old kiln water hazard accident will not occur in the area within the second preset time period, a safety notice message will be pushed to the management personnel.

[0229] Among them, the safety notice message is used to indicate that an old kiln water hazard accident will not occur in the area. After receiving the safety notice message, the management personnel will timely understand the situation of the current area. S802: If it is determined that an old kiln water hazard accident will occur in the area within the second preset time period, a safety warning message will be pushed to the management personnel, where the safety warning message is used to indicate that an old kiln water hazard accident will occur in the area.

[0230] In this step, if it is determined that an old kiln water hazard accident will occur in the area within the second preset time period, a safety warning message will be pushed to the management personnel.

[0231] Specifically, the safety warning information is used to indicate that an old mine water hazard accident will occur in a certain area, aiming to timely remind the management personnel to pay attention to the area where the old mine water hazard accident will occur and take appropriate measures to deal with the possible coal mine water hazard accident. The goal of this warning mechanism is to ensure timely response so that the management personnel can take timely countermeasures, such as increasing the inspection frequency, dispatching coal mine water hazard treatment personnel or reminding users to pay attention to safety, etc. This warning mode helps to improve the efficiency and timeliness of coal mine safety management and reduce the occurrence of coal mine water hazard accidents.

[0232] In a specific implementation manner, the preset working face water inflow data is divided into four parts. Among them, the preset working face water inflow data of the first part is 10m 3 / h, the preset working face water inflow data of the second part is 50m 3 / h, the preset working face water inflow data of the third part is 150m 3 / h, and the preset working face water inflow data of the fourth part is 500m 3 / h.

[0233] If the predicted data of the working face water inflow in the second preset time period is lower than 10m 3 / h, a safety notice information is pushed to the management personnel.

[0234] If the working face water inflow in the second preset time period is higher than or equal to 10m 3 / h and lower than 50m 3 / h, a blue warning information is pushed to the management personnel.

[0235] If the working face water inflow in the second preset time period is higher than or equal to 50m 3 / h and lower than 150m 3 / h, a yellow warning information is pushed to the management personnel.

[0236] If the working face water inflow in the second preset time period is higher than or equal to 150m 3 / h and lower than 500m 3 / h, an orange warning information is pushed to the management personnel.

[0237] If the working face water inflow in the second preset time period is higher than or equal to 500m 3 / h, a red warning information is pushed to the management personnel.

[0238] The intelligent prediction method for coal mine water hazards based on artificial intelligence provided by the embodiments of the present application sends safety notice information or safety warning information to the management personnel according to whether an old mine water hazard accident will occur in the area in the second preset time period. This method can ensure that the management personnel respond in a timely manner and minimize the possible losses and hazards.

[0239] Figure 9 Schematic diagram of a framework of an intelligent early warning system for coal mine water hazards provided for this application. As Figure 9 shown, the intelligent early warning system 900 for coal mine water hazards based on artificial intelligence includes an intelligent perception layer 901, an intelligent control layer 902, an intelligent prediction layer 903, an intelligent early warning layer 904, and an intelligent management layer 905. Among them, when building the intelligent early warning system 900 for coal mine water hazards based on artificial intelligence, an Alibaba Cloud server with 2 cores, 2G memory, 40G cloud disk storage space, and 1Mbps bandwidth is selected to integrate the monitoring data of old goaf water and manage it uniformly. The operating system used is windows11, and a Linux system such as CentOS 8.2 is selected. The processor version is R7-5800H, and the system type is a 64-bit operating system. The compilation tool used is the Spyder editor based on python3.7 and matlab.

[0240] Among them, the intelligent perception layer 901 includes a temperature sensor 9011, a pressure sensor 9012, a water level sensor 9013, a water conduction speed sensor 9014, a pH sensor 9015, an ultrasonic sensor 9016, a water inflow sensor 9017 at the working face, and an ammonia nitrogen compound concentration sensor 9018, which are used to collect the monitoring data of old goaf water in the area to be predicted in the coal mine. At the same time, the intelligent perception layer 901 also includes a dust concentration sensor 9019, a harmful gas concentration sensor 9020, a gas sensor 9021, and a humidity sensor 9022, which are used to collect the working environment data in the area to be predicted in the coal mine.

[0241] Among them, the monitoring data of old goaf water in the area to be predicted in the coal mine can also rely on geological exploration means mainly based on drilling and geophysical prospecting, and use methods such as manual, video, and sensors to obtain real-time monitoring data. By combining and using detection technologies at different stages, real-time, dynamic, and true "four-following" data collection can be achieved.

[0242] The intelligent control layer 902 is mainly used to store the monitoring data of old kiln water and the working environment data of the coal mine area to be predicted transmitted by the intelligent perception layer 901. Specifically, the storage is to transmit the data collected by the intelligent perception layer 901 to the cloud platform. Among them, the transmission methods include but are not limited to direct connection, gateway, wireless communication, and mobile network, etc. Specifically, direct connection means that various sensors in the intelligent perception layer 901 can be directly connected to the cloud server through network interfaces (such as Ethernet, Wi-Fi). The data collected by the sensors can be transmitted to the cloud server through the TCP / IP protocol and processed and stored by the cloud server. Gateway means that the sensors can be connected to a local area network gateway, and the gateway is responsible for transmitting the data of the sensors to the cloud server. The gateway can be connected to the cloud server through Ethernet, Wi-Fi, cellular network, etc., and can use protocols such as MQTT, HTTP, CoAP for data transmission. Wireless communication means that the sensors can use wireless communication technologies (such as Bluetooth, Zigbee, LoRa, etc.) to transmit the data to an intermediate device (such as a mobile phone, embedded system), and then the intermediate device uploads the data to the cloud server. This method is suitable for situations where the distance between the sensors and the cloud server is far or the transmission method is restricted. Mobile network means that the sensors can use the mobile network (such as 4G, 5G) to directly upload the data to the cloud server. This method is suitable for situations where real-time data transmission is required or the sensors need to work in a mobile environment. When choosing the transmission method, it needs to be determined according to the actual situation.

[0243] The intelligent prediction layer 903 is mainly used to predict the predicted water inflow data of the working face within the second preset time period after the current moment. Specifically, the cloud platform system 9031 is deployed in the intelligent prediction layer, and the old kiln water hazard prediction model 9032 is deployed in the cloud platform system 9031. The old kiln water hazard prediction model 9032 is used to predict whether the old kiln water hazard accident will occur. The specific execution process is as follows:

[0244] S90321: Initialization;

[0245] S90322: Detection by multiple sensors;

[0246] Specifically, in this step, it is mainly used to detect the monitoring data of old kiln water in the coal mine area to be predicted.

[0247] S90323: The model detects abnormal data;

[0248] Specifically, in this step, if the model detects abnormal data, step S90321 is executed; if the model does not detect abnormal data, step S90324 is executed.

[0249] S90324: Weighted fusion;

[0250] S90325: The model detects abnormal data;

[0251] Specifically, in this step, if the model detects abnormal data, step S90326 is executed; if the model does not detect abnormal data, step S90321 is executed.

[0252] S90326: Issue an alarm.

[0253] The intelligent early warning layer 904 is mainly used to display safety notification information and safety early warning information, and the display methods include but are not limited to 3D visualization display 9041 and terminal device display 9042, etc. Among them, the terminal device display 9042 includes but is not limited to application software 90421, intelligent applets 90422, and large screens in the monitoring room 90423, etc.

[0254] The intelligent management layer 905 is mainly used for timely intervention 9051, fault analysis 9052, maintenance service 9053, and protection measures 9054.

[0255] Figure 10 It is a schematic flowchart of a coal mine water hazard intelligent prediction method based on artificial intelligence provided by this application. As Figure 10 shown, the process of the coal mine water hazard intelligent prediction method based on artificial intelligence can include:

[0256] S1001: Sensor data training set;

[0257] S1002: Establish a prediction index system;

[0258] S1003: Data division, data standardization;

[0259] S1004: Train the old goaf water hazard prediction model;

[0260] In this step, the training process of the old goaf water hazard prediction model is as follows:

[0261] S10041: Encode the parameters of the coal mine water situation as chromosomes and initialize the population;

[0262] S10042: Calculate the fitness value;

[0263] S10043: Termination condition;

[0264] In this step, if the termination condition is met, step S10045 is executed; if the termination condition is met, step S10044 is executed.

[0265] S10044: Perform genetic operations (selection, crossover, and mutation) to obtain a new population;

[0266] In this step, after obtaining the new population, step S10042 is executed.

[0267] S10045: Decode to obtain the optimal solution;

[0268] S10046: Input the data at time 1 (temperature, pressure, pH, etc.) into LSTM1 to obtain output data 1;

[0269] S10047: Input the data at time 2 (temperature, pressure, pH, etc.) into LSTM1 to obtain output data 2;

[0270] S10048: Input the data at time s (temperature, pressure, pH, etc.) into LSTM1 to obtain output data s;

[0271] S1005: Intelligent prediction of coal mine water hazards;

[0272] S1006: Calculate the error;

[0273] S1007: Output the result.

[0274] Figure 11 This is a schematic structural diagram of Embodiment 1 of an intelligent coal mine water hazard prediction device based on artificial intelligence provided by this application. As Figure 11 shown, the intelligent coal mine water hazard prediction device 1100 based on artificial intelligence includes:

[0275] A receiving module 1101, configured to receive the old goaf water monitoring data of the coal mine's area to be predicted uploaded by the sensing layer device within a first preset time period before the current moment. The old goaf water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old goaf damage monitoring data;

[0276] A processing module 1102, configured to input the old goaf water monitoring data of the area to be predicted within the first preset time period into the old goaf water hazard prediction model for prediction processing to obtain the predicted working face water inrush volume data within a second preset time period after the current moment. The old goaf water hazard prediction model is an LSTM model for predicting the working face water inrush volume that is pre-optimized based on GA, and GA is used to determine the parameters of the LSTM model; the length of the first preset time period is greater than the length of the second preset time period;

[0277] The processing module 1102 is further configured to determine whether an old goaf water hazard accident will occur according to the predicted working face water inrush volume data within the second preset time period of the area to be predicted.

[0278] Optionally, the processing module 1102 is further configured to: determine the target parameter configuration scheme of the LSTM model according to GA;

[0279] Configure the target parameters in the LSTM model according to the target parameter configuration scheme to obtain the old goaf water hazard prediction model.

[0280] Initialize N sets of first parameter configuration schemes, selection operators, adaptive crossover probabilities, adaptive mutation probabilities, and a preset maximum number of iterations for the GA, where N is a positive integer. Each set of first parameter configuration schemes includes the number of hidden layer neurons, learning rate, preset number of iterations, minimum batch size, regularization parameter, weight parameter, and bias term parameter.

[0281] Construct N LSTM models with exactly the same structure to obtain N first LSTM models; for each first LSTM model, select a first chromosome corresponding to the first parameter configuration scheme and configure the initial parameters in the first LSTM model to obtain N second LSTM models, where the initial parameters of the first LSTM model include the initial number of hidden layer neurons, initial learning rate, initial preset number of iterations, initial minimum batch size, initial regularization parameter, initial weight parameter, and initial bias term parameter; select N sets of sample data from the sample data set and input the historical old kiln water monitoring data in the first time period of the N sets of sample data into the N second LSTM models for prediction respectively to obtain the predicted water inflow data of the working face in the second time period; calculate the fitness value of each first chromosome according to the predicted water inflow data of the working face in the N second time periods and the preset fitness function of the GA; sort the fitness values of the N first chromosomes from high to low according to the fitness value of each first chromosome, obtain the first chromosome number corresponding to the highest fitness value, and record it as the target chromosome number; calculate the adaptive crossover probability and adaptive mutation probability of each screened first chromosome according to the fitness value of each screened first chromosome; based on the adaptive crossover probability and adaptive mutation probability of each screened first chromosome, perform adaptive crossover and mutation operations on the screened first chromosomes to generate N second chromosomes; determine whether the current number of iterations has reached the preset maximum number of iterations. If it has not reached the preset maximum number of iterations, determine the N second chromosomes as the new N first chromosomes, and repeat steps e to k until the preset maximum number of iterations is reached. When the preset maximum number of iterations is reached, determine the first parameter configuration scheme corresponding to the obtained target chromosome number as the target parameter configuration scheme.

[0282] Optionally, the processing module 1102 is further configured to: determine, from all the first chromosomes, the first chromosomes whose fitness values meet the preset fitness condition as the first part of chromosomes; determine the second part of chromosomes from all the first chromosomes according to the roulette wheel method. The N screened first chromosomes include the first part of chromosomes and the second part of chromosomes.

[0283] Optionally, the fitness value of each first chromosome is calculated using the following formula:

[0284]

[0285] Among them, represents the fitness value of the nth first chromosome, and f MAE (n) represents the mean absolute error value of the nth first chromosome, and f RMSE (n) represents the root mean square error value of the nth first chromosome.

[0286] Optionally, the processing module 1102 is further configured to: for each screened first chromosome, determine the adaptive crossover probability of the screened first chromosome according to a preset initial crossover probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome;

[0287] Determine the adaptive mutation probability of the screened first chromosome according to a preset initial mutation probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome.

[0288] Optionally, the processing module 1102 is further configured to: if the predicted water inflow data of the working face within the second preset time period is lower than the preset water inflow data of the working face, determine that there will be no old kiln water hazard accident in the area within the second preset time period;

[0289] If the predicted water inflow data of the working face within the second preset time period is higher than or equal to the preset water inflow data of the working face, determine that there will be an old kiln water hazard accident in the area within the second preset time period.

[0290] The intelligent coal mine water hazard prediction device based on artificial intelligence provided in this embodiment can be used to execute the intelligent coal mine water hazard prediction method based on artificial intelligence in any of the foregoing method embodiments. The implementation principle and technical effects are similar and will not be elaborated here.

[0291] Figure 12 This is a schematic structural diagram of Embodiment 2 of an intelligent coal mine water hazard prediction device based on artificial intelligence provided by this application. As Figure 12 shown, the intelligent coal mine water hazard prediction device 1100 based on artificial intelligence further includes:

[0292] An acquisition module 1103, configured to: obtain a sample data set based on historical old kiln water monitoring data. The historical old kiln water monitoring data includes all historical water source monitoring data, historical working face water inrush monitoring data, and historical old kiln water damage monitoring data. The sample data set includes multiple groups of data samples. Each group of data samples includes old kiln water monitoring data for two consecutive time periods, and the duration of the first time period is greater than or equal to the duration of the first preset time period, and the duration of the second time period is the same as the duration of the second preset time period.

[0293] Optionally, the obtaining module 1103 is further configured to: perform chromosome encoding on each set of first parameter configuration schemes to obtain corresponding first chromosomes.

[0294] Optionally, the obtaining module 1103 is further configured to: screen the first chromosomes according to the fitness values corresponding to each first chromosome and the selection operator to obtain N screened first chromosomes.

[0295] The sending module 1104 is configured to: if it is determined that there will be no old kiln water hazard accident in the area within the second preset time period, push a safety notice message to the management personnel, where the safety notice message is used to indicate that there will be no old kiln water hazard accident in the area; if it is determined that there will be an old kiln water hazard accident in the area within the second preset time period, push a safety warning message to the management personnel, where the safety warning message is used to indicate that there will be an old kiln water hazard accident in the area.

[0296] The intelligent coal mine water hazard prediction device based on artificial intelligence provided in this embodiment can be used to execute the intelligent coal mine water hazard prediction method based on artificial intelligence in any of the foregoing method embodiments, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0297] Figure 13 This is a schematic structural diagram of a computer device provided by the present application. As Figure 13 shown, the computer device may specifically include a receiver 1301, a processor 1302, and a memory 1303. Among them, the receiver 1301 is used to realize data transmission between the computer device and the sensing layer device, the memory 1303 stores computer execution instructions; the processor 1302 executes the computer execution instructions stored in the memory 1303 to implement the intelligent coal mine water hazard prediction method based on artificial intelligence in the foregoing embodiment.

[0298] Figure 14 This is an early warning system for coal mine water hazards based on artificial intelligence provided by the present application. As Figure 14 shown, the early warning system 1400 for coal mine water hazards based on artificial intelligence includes a computer device 1401 and a sensing layer device 1402 connected to the computer device. Among them, the sensing layer device 1402 includes: a water source monitoring sensor 14021, a working face water inrush monitoring sensor 14022, and an old kiln damage depth sensor 14023. The water source monitoring sensor 14021 is used to monitor the old kiln water source data; the working face water inrush monitoring sensor 14022 is used to monitor the old kiln water working face water inrush data; the old kiln damage depth sensor 14023 is used to monitor the old kiln damage depth data; the computer device 1401 is used to execute the intelligent coal mine water hazard prediction method based on artificial intelligence in any of the foregoing embodiments.

[0299] The early warning system for coal mine water disasters based on artificial intelligence provided by the embodiments of this application is based on advanced sensing technologies, data collection and analysis technologies, as well as cloud computing and artificial intelligence technologies. Through multiple sensors and monitoring devices arranged in various key areas of the coal mine, it can monitor and sense the water quality, gas concentration, temperature, etc. in real time, and transmit the data to the cloud platform for real-time analysis and processing. At the same time, the LSTM model is adopted, and the GA algorithm is combined to optimize the parameter configuration of the old kiln water disaster prediction model, so as to monitor and accurately predict possible coal mine water disaster accidents in real time for 24 hours. The calculated data is sent to the background through the WIFI / 5G terminal gateway, and the system control device, alarm device and early warning system issue an alarm, so as to achieve the purpose of timely intervention.

[0300] This embodiment provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the artificial-intelligence-based intelligent prediction method for coal mine water disasters in any one of the above embodiments.

[0301] This embodiment also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the artificial-intelligence-based intelligent prediction method for coal mine water disasters provided in any one of the above embodiments.

[0302] Those skilled in the art will readily think of other implementation schemes of this application after considering the specification and practicing the invention disclosed herein. This application aims to cover any variations, uses, or adaptive changes of this application, which follow the general principles of this application and include the common general knowledge or conventional technical means in the technical field not disclosed in this application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of this application are pointed out by the following claims.

[0303] It should be understood that this application is not limited to the exact structure already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is only limited by the appended claims.

Claims

1. An intelligent prediction method for coal mine water hazards based on artificial intelligence, characterized in that, The method includes: Receiving the old kiln water monitoring data of the coal mine's area to be predicted uploaded by the perception layer devices within a first preset time period before the current moment, where the old kiln water monitoring data includes water source monitoring data, working face water inrush monitoring data, and old kiln damage monitoring data; Inputting the old kiln water monitoring data of the area to be predicted within the first preset time period into the old kiln water hazard prediction model for prediction processing to obtain the predicted data of the water inrush volume of the working face within a second preset time period after the current moment, where the old kiln water hazard prediction model is a long short-term memory neural network model LSTM for predicting the water inrush volume of the working face that is pre-optimized based on the genetic algorithm GA, and the GA is used to determine the parameters of the LSTM model; the length of the first preset time period is greater than the length of the second preset time period; Determining whether an old kiln water hazard accident will occur according to the predicted data of the water inrush volume of the working face of the area to be predicted within the second preset time period; The obtaining process of the old kiln water hazard prediction model includes: Determining the target parameter configuration scheme of the LSTM model according to the GA; Configuring the target parameters in the LSTM model according to the target parameter configuration scheme to obtain the old kiln water hazard prediction model; The determining the target parameter configuration scheme of the LSTM model according to the GA includes: Step a: Obtaining a sample data set based on historical old kiln water monitoring data, where the historical old kiln water monitoring data includes all historical water source monitoring data, historical working face water inrush monitoring data, and historical old kiln water damage monitoring data monitored, and the sample data set includes multiple groups of data samples. Each group of data samples includes the old kiln water monitoring data of two consecutive time periods, and the duration of the first time period is greater than or equal to the duration of the first preset time period, and the duration of the second time period is the same as the duration of the second preset time period; Step b: Initializing N groups of first parameter configuration schemes, selection operator, adaptive crossover probability, adaptive mutation probability, and preset maximum iteration times of the GA, where N is a positive integer, and each group of first parameter configuration schemes includes the number of hidden layer neurons, learning rate, preset iteration times, minimum batch processing quantity, regularization parameter, weight parameter, and bias term parameter; Step c: Constructing N LSTM models with exactly the same structure to obtain N first LSTM models; Step d: Performing chromosome coding on each group of first parameter configuration schemes to obtain the corresponding first chromosome; Step e: For each first LSTM model, selecting a first chromosome corresponding to the first parameter configuration scheme and configuring the initial parameters in the first LSTM model to obtain N second LSTM models, where the initial parameters of the first LSTM model include the initial number of hidden layer neurons, initial learning rate, initial preset iteration times, initial minimum batch processing quantity, initial regularization parameter, initial weight parameter, and initial bias term parameter; Step f: Select N groups of sample data from the sample data set, and input the historical old kiln water monitoring data in the first time period of the N groups of sample data into N second LSTM models respectively for prediction to obtain N predicted water inflow data of the working face in the second time period; Step g: Calculate the fitness value of each first chromosome according to the N predicted water inflow data of the working face in the second time period and the preset fitness function of the GA; Step h: Sort the fitness values of the N first chromosomes from high to low according to the fitness value of each first chromosome, obtain the first chromosome number corresponding to the highest fitness value, and record it as the target chromosome number; Step i: Screen the first chromosomes according to the fitness value of each first chromosome and the selection operator to obtain N screened first chromosomes; Step j: Calculate the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome according to the fitness value of each screened first chromosome; Step k: Based on the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome, perform adaptive crossover and mutation operations on the screened first chromosomes to generate N second chromosomes; Step l: Determine whether the current iteration number reaches the preset maximum iteration number. If it does not reach the preset maximum iteration number, determine the N second chromosomes as the new N first chromosomes, and repeat steps e to k until the preset maximum iteration number is reached. When the preset maximum iteration number is reached, determine the first parameter configuration scheme corresponding to the obtained target chromosome number as the target parameter configuration scheme.

2. The method according to claim 1, characterized in that, The screening process of the first chromosomes according to the fitness value of each first chromosome and the selection operator to obtain N screened first chromosomes includes: Among all the first chromosomes, determine the first chromosomes whose fitness values meet the preset fitness conditions as the first part of chromosomes; Determine the second part of chromosomes from all the first chromosomes according to the roulette method. The N screened first chromosomes include the first part of chromosomes and the second part of chromosomes.

3. The method according to claim 1, wherein The fitness value of each first chromosome is calculated using the following formula: Among them, represents the fitness value of the nth first chromosome, f MAE (n) represents the mean absolute error value of the nth first chromosome, f RMSE (n) represents the root mean square error value of the nth first chromosome.

4. The method according to claim 1, characterized in that Calculating the adaptive crossover probability and the adaptive mutation probability of each screened first chromosome according to the fitness value of each screened first chromosome includes: For each screened first chromosome, determine the adaptive crossover probability of the screened first chromosome according to the preset initial crossover probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome; Determine the adaptive mutation probability of the screened first chromosome according to the preset initial mutation probability, the current iteration number, the preset maximum iteration number, and the fitness of the screened first chromosome.

5. The method according to claim 1, characterized in that, Determining whether an old kiln water hazard accident will occur according to the predicted water inflow data of the working face in the second preset time period in the area to be predicted, the method includes: If the predicted water inflow data of the working face within the second preset time period is lower than the preset water inflow data of the working face, it is determined that there will be no old kiln water hazard accident in the area within the second preset time period; If the predicted water inflow data of the working face within the second preset time period is higher than or equal to the preset water inflow data of the working face, it is determined that there will be an old kiln water hazard accident in the area within the second preset time period.

6. The method according to claim 5, characterized in that, The method further includes: If it is determined that there will be no old kiln water hazard accident in the area within the second preset time period, a safety notification message is pushed to the management personnel, where the safety notification message is used to indicate that there will be no old kiln water hazard accident in the area; If it is determined that there will be an old kiln water hazard accident in the area within the second preset time period, a safety warning message is pushed to the management personnel, where the safety warning message is used to indicate that there will be an old kiln water hazard accident in the area.

7. A computer device, characterized in that, Comprising: A receiver, a processor, and a memory connected to the processor; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory to implement the artificial intelligence-based intelligent prediction method for coal mine water hazards according to any one of claims 1 to 6.

8. An early warning system for coal mine water disasters based on artificial intelligence, characterized in that, Comprising: A computer device and a perception layer device connected to the computer device; The perception layer device includes: a water source monitoring sensor, a working face water inflow monitoring sensor, and an old kiln damage depth sensor; The water source monitoring sensor is used to monitor the old kiln water source data; The working face water inflow monitoring sensor is used to monitor the old kiln water working face water inflow data; The old kiln damage depth sensor is used to monitor the old kiln damage depth data; The computer device is used to execute the artificial intelligence-based intelligent prediction method for coal mine water hazards according to any one of claims 1 to 6.

Images