Intelligent regulation and control method for coal mine ventilation system based on environmental perception

Abstract

The invention discloses a coal mine ventilation system intelligent regulation and control method based on environmental perception, and relates to the technical field of coal mine ventilation system intelligent regulation and control, and the method comprises the following steps: collecting environmental data of different areas of a mine, including gas concentration, temperature and humidity, air velocity and oxygen content, and monitoring the operation state of ventilation equipment in real time; data of all the sensors are synchronously processed through the data fusion technology, and a unified environment sensing model is generated. According to the invention, through cooperative work of the expert system and deep learning, real-time monitoring and emergency plan execution under extreme working conditions are realized, and the response speed and decision accuracy are improved. A reinforcement learning algorithm is adopted to optimize ventilation line switching, the adaptive capacity and redundancy guarantee of the system are enhanced, and it is ensured that airflow is not interrupted at the critical moment. Through abnormality comparison of the knowledge base and the rule base, false alarm and misjudgment are reduced, system efficiency is improved, operation burden is relieved, and safety and stable operation of mine personnel are guaranteed to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of coal mine ventilation systems, and in particular to an intelligent control method of coal mine ventilation systems based on environmental perception. Background Art

[0002] Intelligent control of coal mine ventilation systems based on environmental perception refers to the dynamic adjustment of ventilation systems in coal mines through the integration of environmental monitoring sensors and intelligent algorithms. The system can sense key environmental parameters inside the mine in real time, such as gas concentration, temperature and humidity, wind speed, and dust level, and optimize the control of air volume and direction through data analysis and prediction models. Intelligent control can not only automatically increase ventilation according to changes in hazardous gases, but also reduce ventilation when it is not necessary, thereby reducing energy consumption and operating costs, ensuring the safety of miners and the controllability of the environment. Such a system effectively improves the accuracy and efficiency of coal mine ventilation management, reduces manual intervention, and improves the safety and environmental protection level of mines.

[0003] The prior art has the following deficiencies:

[0004] When the coal mine ventilation system is intelligently controlled, in complex mine environments, such as extreme situations such as sudden gas eruptions, fires, or local collapses, the system's intelligent algorithm may cause response delays or misjudgments due to lack of sufficient training data or the model's over-reliance on normal operating data. For example: if the system fails to adjust the ventilation volume or switch to backup ventilation lines in time, harmful gases in the dangerous area cannot be quickly discharged or oxygen supply is insufficient. This lag may cause major safety accidents, such as gas explosions or worker suffocation.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent control method for coal mine ventilation system based on environmental perception. Through the collaborative work of expert system and deep learning, the response lag problem of traditional ventilation system under extreme working conditions is solved, and the generation of emergency plan in millisecond level is realized to ensure accurate decision-making and reduce the risk of gas explosion and oxygen shortage. With the help of virtual simulation training of reinforcement learning algorithm, the system can adaptively select the optimal ventilation path when sudden collapse or air duct is blocked, and ensure the stability of airflow through multi-line exhaust, thereby improving the reliability of the system. Through the abnormal comparison and analysis of knowledge base and rule base, the sensor false alarm is effectively reduced, the system efficiency is improved, and the burden on operators is reduced, so as to maximize the protection of mine safety and the stability of production environment, so as to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: an intelligent control method for coal mine ventilation system based on environmental perception, which dynamically monitors and controls the complex environment of coal mines in real time by combining multi-source environmental data collection with intelligent algorithms to cope with the response delay and misjudgment problems under extreme working conditions, and specifically includes the following steps:

[0008] Collect environmental data from different areas of the mine, including gas concentration, temperature and humidity, air flow rate and oxygen content, and monitor the operating status of ventilation equipment in real time; synchronize the data of each sensor through data fusion technology to generate a unified environmental perception model to ensure the consistency and accuracy of data in different areas inside and outside the mine;

[0009] Use a multi-level early warning model to identify potential extreme situations. Use a threshold detection algorithm based on historical data and a rule base to determine whether extreme conditions exist. In the event of an extreme condition, send a real-time alert to management personnel and trigger the system's automatic response module.

[0010] The prediction model of air flow and gas concentration changes in the mine is continuously updated through the adaptive neural network algorithm. When a sharp fluctuation in environmental data is detected, the neural network algorithm immediately calculates the optimal ventilation adjustment strategy and sends instructions to related fan equipment to ensure that harmful gases are quickly discharged.

[0011] Adopting a reinforcement learning-based algorithm, the system pre-trains ventilation line switching schemes for various extreme situations in a virtual simulation environment. When the main ventilation line is blocked, the system automatically selects the optimal backup ventilation line based on the current working conditions to ensure that the airflow in key areas is not interrupted.

[0012] By introducing an expert system and combining the knowledge base with a deep learning algorithm, the nonlinear relationship between multiple variables in complex situations can be analyzed. When the system algorithm encounters uncertain situations or data anomalies, the expert system is triggered to compare and analyze historical cases to assist in making key decisions and ensure that the algorithm has redundant safety protection when executing strategies.

[0013] Preferably, the detailed steps of synchronously processing sensor data and generating a unified environment perception model through data fusion technology are as follows:

[0014] First, real-time data is collected from a variety of sensors placed in different areas of the mine;

[0015] Perform time synchronization and space alignment processing;

[0016] The data from different types of sensors are fused and processed, and a multi-source information fusion algorithm is used to unify the data from multiple sensors into an environmental perception model;

[0017] After completing data fusion, the system generates a unified environmental perception model inside and outside the mine and updates it dynamically in real time.

[0018] Preferably, the detailed steps of the multi-level warning model to identify potential extreme situations and trigger the automatic response module are as follows:

[0019] The system first obtains environmental data from sensors in the mine and compares the acquired environmental data with historical data; the multi-level early warning model gradually identifies potential extreme conditions by setting multiple risk levels, and each level corresponds to a different value range and response strategy;

[0020] Using the threshold detection algorithm based on historical data and rule base, real-time analysis is performed to determine whether the collected environmental data has reached the critical value of extreme working conditions;

[0021] Once the system detects an extreme working condition, it immediately triggers the preset alarm system and notifies the mine management personnel through multiple channels. It also generates a real-time alarm report based on the mine emergency plan and records the entire event detection and response process through the log function to facilitate subsequent analysis and optimization.

[0022] When extreme working conditions are detected, the preset emergency module is automatically triggered.

[0023] Preferably, the detailed steps of updating the prediction model of air flow and gas concentration in the mine and optimizing the ventilation volume based on the adaptive neural network algorithm are as follows:

[0024] First, the neural network model receives real-time environmental data collected by sensors as input, including gas concentration C(t), temperature T(t), humidity H(t), air flow rate v(t) and oxygen content O2(t), and the input data is composed of a vector X(t) = [C(t), T(t), H(t), v(t), O2(t)] and input into the adaptive neural network model in real time. The neural network uses a recursive neural network or a long short-term memory network to capture the complex time series relationship in the mine, and uses the following formula to update the internal state of the model: h t =σ(W h ·h t-1 +W x ·X(t)+b h ), where h t is the hidden layer state, W h and W x are the weights of the state transfer matrix and the input matrix, respectively, h is the bias vector, σ is the activation function, and its internal state is gradually updated by continuously inputting time series data, providing a basis for the subsequent prediction and optimization of ventilation volume;

[0025] Through the training of adaptive neural networks, the changing rules of air flow and gas concentration in the mine are continuously learned, and the prediction model is updated in real time. The prediction model is based on a multi-step prediction strategy to calculate the gas concentration and air flow rate at the future time t+Δt. The prediction process is expressed by the following formula: in, and are the predicted future gas concentration and air flow rate respectively; f and g are the nonlinear mapping functions of the neural network; ∈ and η are noise terms used to simulate random interference factors in the mine.

[0026] Preferably, when a sharp fluctuation in the mine environment data is detected, the neural network immediately calculates the optimal ventilation adjustment strategy to achieve rapid discharge of harmful gases. The optimization strategy is solved based on the following objective function:

[0027]

[0028] , where Q(t) is the ventilation volume of the fan at time t; C safe is the gas concentration under the safety threshold; v target is the target air flow rate; E(Q) is the energy consumption function of the ventilation system; α, β, γ are weight parameters that control the balance between safety, flow rate and energy consumption.

[0029] Preferably, the detailed steps of virtual simulation training based on reinforcement learning algorithm and optimal ventilation line switching are as follows:

[0030] First, a virtual simulation environment is built to simulate the actual terrain structure and ventilation system of the mine, including the layout of the air ducts, the distribution of the main and backup fans, and the key operating areas of the mine;

[0031] Using a reinforcement learning framework, the mine ventilation system is defined as the "environment" and the algorithm as the "agent". The algorithm learns how to make optimal decisions under different extreme working conditions by constantly interacting with the virtual environment.

[0032] After a lot of training, the reinforcement learning algorithm forms a policy network to select the optimal ventilation path in different scenarios;

[0033] When the main ventilation line is blocked, the system immediately calls the optimal backup ventilation plan calculated by the reinforcement learning algorithm and automatically performs the switching operation. The algorithm will quickly select a path to ensure uninterrupted airflow in key areas based on current sensor data and status.

[0034] Preferably, the specific steps for implementing decision support by combining an expert system with a deep learning algorithm are as follows:

[0035] Deep learning models are introduced to handle complex nonlinear relationships. Through training with a large amount of mine data, deep learning models can predict the changing trends and potential risks between environmental variables. The framework combines expert systems and deep learning. When the system receives new data, it triggers corresponding expert rules based on the prediction results of deep learning.

[0036] Continuously monitor the mine's environmental data and trigger the anomaly detection module when data anomalies are detected;

[0037] After receiving an abnormal situation, the expert system selects the case that best fits the current situation from the knowledge base through the reasoning engine and compares it with the prediction results of the deep learning model;

[0038] After completing the joint analysis, the system generates a specific emergency decision-making plan and provides decision-making suggestions to mine managers through a visual interface, while indicating potential risks and implementation steps.

[0039] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0040] The present invention effectively solves the problem of delayed response of traditional ventilation systems under extreme working conditions by introducing the collaborative work of expert systems and deep learning. When an emergency such as a fire, gas eruption or landslide occurs, the system uses a deep learning model to conduct real-time monitoring and prediction, quickly analyze the complex changes inside the mine, and automatically trigger the emergency plan in the expert system. This collaborative mechanism enables the system to generate emergency plans within milliseconds, reducing the time delay caused by relying on manual judgment. Unlike traditional fixed strategies, the present invention ensures the accuracy and adaptability of decision-making based on intelligent reasoning of historical cases and real-time data. This efficient and accurate response capability greatly reduces the risks of gas explosions, lack of oxygen, etc., and provides a strong guarantee for the safe operation of mines.

[0041] The present invention adopts virtual simulation training of reinforcement learning algorithm, so that the system can optimize the switching strategy of ventilation lines under various complex conditions. In actual operation, when a main air duct is blocked due to landslide or equipment failure, the system automatically selects the optimal backup ventilation path based on the adaptive model of reinforcement learning, and adjusts the fan output and air duct status according to the real-time working conditions. At the same time, the system has a built-in redundant safety mechanism, which allows the airflow to be maintained stable through the joint exhaust of multiple lines when one line fails. This dynamic adaptive capability ensures the high reliability of the system in the face of uncertainty and emergencies, and greatly reduces the potential accidents caused by air duct blockage or insufficient ventilation.

[0042] The present invention significantly reduces the risk of misjudgment caused by sensor false alarms and data anomalies through abnormal comparison analysis of the knowledge base and the rule base, effectively avoids redundant alarms triggered by single data errors, and improves the operating efficiency of the system. At the same time, the system also provides detailed risk assessment and emergency decision support through joint analysis of expert systems and deep learning, reducing the judgment burden of operators in complex situations, not only improving the operational efficiency of the ventilation system, but also ensuring the life safety of personnel in the mine and the stability of the production environment to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0044] Figure 1 This is a method flow chart of an intelligent control method for a coal mine ventilation system based on environmental perception according to the present invention. DETAILED DESCRIPTION

[0045] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0046] The present invention provides Figure 1 The intelligent control method of coal mine ventilation system based on environmental perception is shown. By combining multi-source environmental data collection with intelligent algorithms, the complex environment of the coal mine is dynamically monitored and controlled in real time to cope with the response delay and misjudgment problems under extreme working conditions. The method specifically includes the following steps:

[0047] Collect environmental data from different areas of the mine, including gas concentration, temperature and humidity, air flow rate and oxygen content, and monitor the operating status of major ventilation equipment in real time; synchronize the data of each sensor through data fusion technology to generate a unified environmental perception model to ensure data consistency and accuracy in different areas inside and outside the mine;

[0048] The detailed steps of synchronously processing sensor data and generating a unified environmental perception model through data fusion technology are as follows:

[0049] First, real-time data is collected from a variety of sensors (such as gas concentration sensors, temperature and humidity sensors, air flow sensors, and oxygen detection sensors) arranged in different areas of the mine. Since these sensors may be located at different depths, areas, or near equipment, the acquisition frequency and data format are different, so preprocessing is necessary. Preprocessing includes standardization of data formats (such as uniform conversion of different units), outlier filtering (eliminating sudden erroneous data), and filling of missing data (through linear interpolation or nearest value filling) to ensure the reliability and consistency of the data source.

[0050] The mine environment is complex, and there may be deviations in the collection time of sensors in different areas, so time synchronization and spatial alignment are required. Time synchronization ensures that the measurement data of different sensors are based on the same time scale by using the Network Time Protocol (NTP) or internal timestamps. Spatial alignment is performed based on the geographical location and installation location of the sensor to ensure that the data corresponds to the specific area or ventilation path in the mine in space. This process helps to avoid misjudgment due to data deviation in time or space, making subsequent analysis more accurate.

[0051] The data from different types of sensors are fused and processed, and multi-source information fusion algorithms such as Kalman filtering, Bayesian estimation or weighted average fusion methods are used to unify the data from multiple sensors into an environmental perception model. In this process, the data needs to be processed differently according to the weight, accuracy and real-time performance of different sensors. For example, the weight of gas concentration sensors in key areas may be higher, while air flow rate data needs to be updated frequently to reflect changes in wind direction in the mine. The fused data not only provides a more comprehensive environmental status, but also reduces the error caused by the failure of a single sensor.

[0052] After completing the data fusion, the system generates a unified environmental perception model inside and outside the mine, and updates it dynamically in real time. The model displays the environmental status of each area of ​​the mine through a visual interface, including gas concentration distribution, temperature and humidity conditions, oxygen content, and wind speed and direction. As new data is continuously input, the system automatically updates the model to ensure that managers can obtain the latest mine environmental information and monitor the ventilation system in real time. In addition, the accumulation of historical data in the model also provides support for the training of intelligent algorithms and the prediction of extreme working conditions, further improving the system's prediction capabilities and emergency response efficiency.

[0053] Use multi-level early warning models to identify potential extreme situations (such as a sharp increase in gas concentration or abnormal temperature rise); use the threshold detection algorithm established by historical data and rule base to determine whether extreme working conditions exist, and send real-time alarms to management personnel in extreme working conditions, while triggering the system's automatic response module;

[0054] The detailed steps of the multi-level early warning model to identify potential extreme situations and trigger the automatic response module are as follows:

[0055] The system first obtains environmental data such as gas concentration, temperature, humidity, oxygen content and air flow rate from sensors in the mine, and compares these data with historical data. The multi-level warning model gradually identifies potential extreme conditions by setting multiple risk levels, such as normal, warning, alarm and emergency. Each level corresponds to a different numerical range and response strategy. For example, when the gas concentration is slightly higher than the normal value, the system enters the warning state; if it continues to rise, a higher level of alarm is triggered. Such a hierarchical mechanism ensures that the system can identify anomalies early and reduce the probability of sudden accidents.

[0056] The system uses a threshold detection algorithm based on historical data and a rule base to analyze in real time whether the collected environmental data has reached the critical value of extreme working conditions. The rule base is based on previous accident cases and safety regulations in the mine, including a combination of multi-dimensional conditions, such as gas concentration continuously exceeding a certain level, sudden drop in air flow rate, or large temperature fluctuations in a short period of time. The algorithm not only supports the monitoring of a single variable, but also detects complex hidden dangers through multi-parameter correlation analysis. For example, high gas concentration accompanied by a drop in air flow rate will be identified as a higher risk. If it is detected that the preset threshold is exceeded, the system will immediately upgrade to a higher level of alarm.

[0057] Once the system detects an extreme condition, it immediately triggers the preset alarm system and notifies mine managers through multiple channels, such as text messages, voice alarms, and emergency notifications on the control room display. The alarm not only indicates the specific risk type (such as excessive gas or fire), but also indicates the area of ​​occurrence and the recommended emergency measures. Managers can view detailed data trends and the response strategies recommended by the system through a visual interface to ensure quick decision-making. In addition, the system will generate a real-time alarm report based on the mine emergency plan, and record the entire event detection and response process through the log function to facilitate subsequent analysis and optimization.

[0058] After detecting extreme working conditions, the system not only relies on manual response, but also automatically triggers the preset emergency module. For example, when the gas concentration exceeds the safety threshold and continues to rise in a short period of time, the system will automatically adjust the speed of the main fan or enable the backup ventilation line to exhaust harmful gases and increase oxygen supply. At the same time, if the risk of fire or landslide is detected, the system will activate the corresponding isolation device or close some air ducts to prevent the spread of danger. The entire process continuously monitors the execution effect through the feedback loop, and dynamically adjusts the control strategy to ensure the efficient implementation of emergency measures and minimize the occurrence of safety accidents.

[0059] The prediction model of air flow and gas concentration changes in the mine is continuously updated through the adaptive neural network algorithm. When a sharp fluctuation in environmental data is detected, the neural network algorithm immediately calculates the optimal ventilation adjustment strategy and sends instructions to relevant fan equipment to ensure that harmful gases are quickly discharged.

[0060] The detailed steps of updating the prediction model of air flow and gas concentration in the mine and optimizing the ventilation volume based on the adaptive neural network algorithm are as follows:

[0061] First, the neural network model receives real-time environmental data collected by sensors as input, including gas concentration C(t), temperature T(t), humidity H(t), air flow rate v(t) and oxygen content O2(t). These input data form a vector X(t) = [C(t), T(t), H(t), v(t), O2(t)], which is input into the adaptive neural network model in real time. The neural network uses a recurrent neural network (RNN) or a long short-term memory network (LSTM) to capture the complex time series relationship in the mine and update the internal state of the model using the following formula: h t =σ(W h ·h t-1 +W x ·X(t)+b h ), where h t is the hidden layer state, W h and W x are the weights of the state transfer matrix and the input matrix, respectively, h is the bias vector, and σ is the activation function (such as ReLU or Tanh). By continuously inputting time series data, the model gradually updates its internal state, providing a basis for the subsequent prediction and optimization of ventilation volume.

[0062] Through the training of adaptive neural networks, the system continuously learns the changing patterns of air flow and gas concentration in the mine and updates the prediction model in real time. The prediction model is based on a multi-step prediction strategy to calculate the gas concentration and air flow rate at the future time t+Δt. The prediction process is expressed by the following formula: in, and are the predicted future gas concentration and air flow rate, respectively; f and g are the nonlinear mapping functions of the neural network; ∈ and η are noise terms used to simulate random interference factors in the mine. After each new data input, the model adjusts the weights through the back propagation algorithm (BPTT) and the gradient descent method to adaptively optimize the prediction accuracy.

[0063] When a sharp fluctuation in the mine environment data is detected (such as a rapid rise in C(t) or a sudden drop in v(t)), the neural network immediately calculates the optimal ventilation adjustment strategy to achieve rapid discharge of harmful gases. The optimization strategy is solved based on the following objective function: Where, Q(t) is the ventilation volume of the fan at time t; C safe is the gas concentration under the safety threshold; v target is the target air flow rate; E(Q) is the energy consumption function of the ventilation system; α, β, γ are weight parameters that control the balance between safety, flow rate and energy consumption. After solving the optimization problem by gradient descent or Lagrange multiplier method, the system immediately sends instructions to the fan equipment to adjust the ventilation volume to ensure that harmful gases are quickly discharged and the mine environment is restored to a safe state.

[0064] Adopting a reinforcement learning-based algorithm, the ventilation line switching schemes for various extreme situations are pre-trained in a virtual simulation environment. When a fire, landslide, etc. cause some main ventilation lines to be blocked, the algorithm automatically selects the optimal backup ventilation line according to the current working conditions to ensure that the airflow in key areas is not interrupted.

[0065] The detailed steps of virtual simulation training and optimal ventilation route switching based on reinforcement learning algorithm are as follows:

[0066] First, a virtual simulation environment is built to simulate the real terrain structure and ventilation system of the mine, including the layout of the air ducts, the distribution of the main and backup fans, and the key operating areas of the mine. The virtual environment also needs to accurately simulate the impact of various extreme working conditions (such as fire, gas eruption, landslide, etc.) on the ventilation path. The virtual sensors configured in the simulation environment can simulate the real-time changes in gas concentration, wind speed and temperature, providing dynamic virtual data input for the algorithm. Through this high-precision simulation system, the reinforcement learning algorithm can be repeatedly trained in a variety of extreme scenarios to explore and optimize the ventilation line switching scheme.

[0067] Using a reinforcement learning framework (such as deep Q-network (DQN) or proximal policy optimization (PPO)), the mine ventilation system is defined as the "environment" and the algorithm as the "agent". The algorithm learns how to make optimal decisions under different extreme working conditions by constantly interacting with the virtual environment. During each simulation run, the algorithm takes corresponding ventilation line switching actions based on the current working conditions (such as blocked air ducts or abnormal gas concentrations), and receives rewards or penalties based on the results. For example, if the switched line successfully exhausts harmful gases and maintains oxygen supply, the algorithm is rewarded; conversely, if the decision causes airflow obstruction, a penalty is imposed. Through multiple training sessions, the algorithm gradually optimizes the selection strategy and achieves the ability to respond quickly to different complex situations.

[0068] After a lot of training, the reinforcement learning algorithm will form a policy network for selecting the optimal ventilation path under different scenarios. However, since the mine environment may change over time (such as the opening of new air ducts or the blockage of existing air ducts), the system must support dynamic policy updates. In actual operation, the algorithm will continue to collect real-time data and compare it with the results in the virtual simulation. If the current working conditions deviate from the training data, the system will trigger new virtual training and update the policy network. In this way, the algorithm not only adapts to static historical data, but also adaptively optimizes the ventilation route according to the changes in the mine.

[0069] When a fire, landslide, etc. occurs in the mine and some main ventilation lines are blocked, the system will immediately call the optimal backup ventilation plan calculated by the reinforcement learning algorithm and automatically perform the switching operation. Based on the current sensor data and status, the algorithm will quickly select a path to ensure that the airflow in key areas is not interrupted. For example, if a main air duct is blocked by a landslide, the system will automatically start the backup air duct and fan and adjust the air volume distribution to maintain air circulation in each area. The entire switching process is monitored by a feedback loop to ensure that the switched line can operate normally. If there is an abnormality, the system will automatically recalculate the alternative plan to further improve the efficiency and reliability of emergency response.

[0070] The system introduces an expert system, which analyzes the nonlinear relationship between multiple variables in complex situations by combining the knowledge base with deep learning algorithms. When the system algorithm encounters uncertain situations or data anomalies, the expert system is triggered to compare and analyze historical cases to assist in making key decisions and ensure that the algorithm has redundant security when executing strategies.

[0071] The specific steps for implementing decision support combining expert system and deep learning algorithm are as follows:

[0072] The core of the expert system is the knowledge base, which contains the operating rules, safety regulations and historical case data of the mine ventilation system. The rules in the knowledge base are structured through a chain of logical reasoning, such as specific plans and emergency plans for dealing with situations such as excessive gas, fire, and landslides. At the same time, the system introduces deep learning models (such as LSTM or Transformer networks) to handle complex nonlinear relationships. These models are trained with a large amount of mine data and can predict the changing trends and potential risks between environmental variables. The framework combines expert systems and deep learning. When the system receives new data, it triggers the corresponding expert rules based on the prediction results of deep learning, thereby improving the intelligence level of emergency response.

[0073] The system continuously monitors the mine's environmental data, such as gas concentration, temperature, humidity, and wind speed. When data anomalies are detected (such as data fluctuations beyond the expected range or inconsistent with model predictions), the anomaly detection module is triggered. For example, if the gas concentration rises sharply in a short period of time but the fan operates normally, it means that there may be unforeseen pipe blockages or sensor errors. At this time, the system will call the historical case library in the expert system to compare with the latest predictions of the deep learning model, analyze whether there are similar situations, and determine whether the current working conditions require manual intervention or enter the automatic processing process.

[0074] After receiving an abnormal situation, the expert system uses the reasoning engine to select the case closest to the current situation from the knowledge base and compare it with the prediction results of the deep learning model. For example, if an abnormal gas concentration is detected and the data shows that the main ventilation line is blocked, the system will infer that a landslide may have occurred and recommend switching to the backup air duct. The reasoning results of the expert system complement the analysis of deep learning to avoid misjudgment of a single algorithm. Joint analysis not only improves the accuracy of judgment, but also provides redundancy for the system's automatic response, ensuring that even if some modules fail, the system can still make reasonable countermeasures.

[0075] After completing the joint analysis, the system generates a specific emergency decision plan and provides decision-making suggestions to mine managers through a visual interface, while indicating potential risks and execution steps. For example, the system will recommend a specific path for switching to a backup ventilation line and inform the possible impact area and time window. If the manager confirms the execution, the system will automatically implement the decision and continuously monitor the actual effect during the execution process. If the actual situation does not meet expectations, the system will quickly trigger the redundant safety mechanism of the expert system, re-call the historical case library and deep learning model for analysis, and ensure that the system has the ability to continuously adjust to maximize mine safety.

[0076] Specific implementation method 1: Automatic judgment and emergency handling of abnormal situations based on knowledge base;

[0077] In the daily operation of the mine, due to fluctuations in sensor data or sudden changes in the environment, key parameters such as gas concentration, temperature, humidity, and wind speed may be abnormal. With the help of the expert system's knowledge base, the system realizes automatic judgment and emergency handling of these abnormal situations. When the system detects that the gas concentration rises rapidly in a short period of time, but the operating status of the main ventilation equipment is normal, the system triggers the expert system's knowledge base comparison module. A large number of historical cases and operating rules are pre-stored in the knowledge base, covering scenarios such as sensor false alarms, air duct blockages, and ventilation equipment failures. The system will quickly compare with these historical cases to determine whether the current abnormality is caused by equipment failure, false alarms, or changes in actual working conditions.

[0078] If the comparison finds that the current situation is highly similar to a recorded false alarm event, for example, a sensor has sent erroneous data due to a fault, the system will promptly prompt the manager to check the sensor status and switch to the data source of the backup sensor. At the same time, the system will analyze the associated data of other sensors, such as oxygen content and air flow rate, to further verify whether the abnormal gas concentration is a sensor error. If the system determines it to be a false alarm, it will automatically mark the data as abnormal and prevent it from entering the further decision-making process, avoiding the system triggering unnecessary emergency responses.

[0079] In more complex situations, such as when the gas concentration increases with the wind speed of the main ventilation duct decreases, the system will determine that there may be air duct blockage. At this time, the expert system will generate emergency suggestions based on historical blockage cases in the knowledge base, prompting managers to conduct on-site inspections, and simulate whether it is necessary to increase ventilation or switch to backup air ducts within the system. If it is judged to be a high-risk event, such as the gas concentration continues to rise and the air duct resistance has not decreased, the system will upgrade the alarm and recommend immediate isolation and evacuation operations. This implementation method reduces the time of manual judgment and improves the response speed of mine environmental management through the rapid comparison and emergency processing of the expert system, ensuring the safety and reliability of operation.

[0080] Specific implementation method 2: Deep learning prediction model assists ventilation line switching optimization;

[0081] When a local collapse occurs in the mine and some main air ducts are blocked, the ventilation system must quickly switch to backup lines to maintain smooth airflow. To this end, the system introduces a deep learning prediction model to analyze real-time environmental data and historical records to predict the changing trends of variables such as gas concentration, wind speed, temperature and humidity, so as to select the optimal ventilation path. The system has simulated and trained different scenarios of the mine in advance, and has mastered the performance of each ventilation line under different conditions through a deep learning model. The model can calculate the airflow capacity of the backup line in various situations in real time, and estimate the gas emission effect and oxygen supply level in advance.

[0082] For example, when a main air duct in a mine is blocked by a landslide, the deep learning model will take over the analysis of the current data and perform simulation calculations on each backup line. If the wind speed of a backup line is not enough to remove the gas in a short time, the system will continue to evaluate other lines and comprehensively consider whether to enable a multi-line joint exhaust strategy. In this case, the system may enable two or more backup air ducts at the same time to avoid the problem of insufficient flow when a single line is switched. In addition, the model also predicts the impact of airflow switching on surrounding areas, such as increased wind speed may cause increased gas concentration or temperature fluctuations in other areas. Through multi-line joint exhaust, the system achieves flexible scheduling in complex scenarios, ensuring that air circulation in the mine can be maintained in extreme situations such as landslides or air duct blockages.

[0083] During the whole process, the system will also monitor the execution effect after the switch in real time, and compare the actual data with the model prediction. If it is found that the performance after the air duct switch does not meet expectations (such as the gas concentration does not drop as predicted), the system will immediately recalculate the backup plan and make secondary adjustments. This method ensures the adaptability of the ventilation system in a dynamic environment and minimizes the risk of airflow obstruction and oxygen deficiency through the optimization capabilities of the deep learning algorithm.

[0084] Specific implementation method three: Collaborative emergency plan execution of expert system and deep learning;

[0085] When extreme situations such as fire and gas eruption occur in the mine, the system realizes accurate execution of emergency plans through the collaboration of expert system and deep learning. The knowledge base of the expert system stores detailed safety regulations and emergency plans, including how to switch ventilation lines, how to isolate dangerous areas, and the steps for personnel evacuation. When extreme situations occur, the deep learning model will first quickly analyze the current environmental variables, such as a sharp increase in temperature and an abnormal increase in gas concentration, and predict the impact of these changes on various areas. Based on these prediction results, the system will send a request to the expert system to call the most appropriate emergency plan.

[0086] For example, when the deep learning model predicts that the high temperature caused by the fire may damage the main air duct, the expert system will refer to the fire plan in the knowledge base and recommend closing the affected area and switching to the backup air duct. At the same time, the system will automatically increase the output power of the main fan and the backup fan to ensure the efficiency of emergency ventilation. Throughout the process, the deep learning model will continuously provide real-time data support to ensure that the expert system's plan execution is in line with the current actual situation. If the switched air duct fails to achieve the expected effect, the system will trigger a new calculation cycle and dynamically adjust the emergency plan.

[0087] The advantage of this collaborative approach is that it improves the system's emergency response speed under complex working conditions. The deep learning model is responsible for analyzing complex nonlinear relationships and real-time predictions, while the expert system provides the execution steps of the emergency plan through the knowledge base. The combination of the two not only improves the accuracy of decision-making, but also effectively avoids misjudgment or lag in a single system. Through this collaborative mechanism, the system has a high degree of adaptability and flexibility, and can ensure the safe operation of the mine and the timely evacuation of personnel even in complex and extreme environments.

[0088] The present invention effectively solves the problem of delayed response of traditional ventilation systems under extreme working conditions by introducing the collaborative work of expert systems and deep learning. When an emergency such as a fire, gas eruption or landslide occurs, the system uses a deep learning model to conduct real-time monitoring and prediction, quickly analyze the complex changes inside the mine, and automatically trigger the emergency plan in the expert system. This collaborative mechanism enables the system to generate emergency plans within milliseconds, reducing the time delay caused by relying on manual judgment. Unlike traditional fixed strategies, the present invention ensures the accuracy and adaptability of decision-making based on intelligent reasoning of historical cases and real-time data. This efficient and accurate response capability greatly reduces the risks of gas explosions, lack of oxygen, etc., and provides a strong guarantee for the safe operation of mines.

[0089] The present invention adopts virtual simulation training of reinforcement learning algorithm, so that the system can optimize the switching strategy of ventilation lines under various complex conditions. In actual operation, when a main air duct is blocked due to landslide or equipment failure, the system automatically selects the optimal backup ventilation path based on the adaptive model of reinforcement learning, and adjusts the fan output and air duct status according to the real-time working conditions. At the same time, the system has a built-in redundant safety mechanism, which allows the airflow to be maintained stable through the joint exhaust of multiple lines when one line fails. This dynamic adaptive capability ensures the high reliability of the system in the face of uncertainty and emergencies, and greatly reduces the potential accidents caused by air duct blockage or insufficient ventilation.

[0090] The present invention significantly reduces the risk of misjudgment caused by sensor false alarms and data anomalies through abnormal comparison and analysis of the knowledge base and the rule base. For example, when the sensor falsely reports that the gas concentration is too high, the system will use historical cases and other sensor data for comparison and analysis to confirm whether further emergency measures are needed. This mechanism avoids redundant alarms triggered by single data errors and improves the operating efficiency of the system. At the same time, the system also provides detailed risk assessment and emergency decision support through the joint analysis of expert systems and deep learning, reducing the judgment burden of operators in complex situations. This intelligent processing method not only improves the operational efficiency of the ventilation system, but also maximizes the safety of life of personnel in the mine and the stability of the production environment.

[0091] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent control method for a coal mine ventilation system based on environmental perception, characterized in that: By combining multi-source environmental data collection with intelligent algorithms, dynamic monitoring and real-time regulation of the complex environment of coal mines are carried out to deal with response delays and misjudgment problems under extreme working conditions. The specific steps include: Collect environmental data from different areas of the mine, including gas concentration, temperature and humidity, air flow rate and oxygen content, and monitor the operating status of ventilation equipment in real time; synchronize the data of each sensor through data fusion technology to generate a unified environmental perception model to ensure the consistency and accuracy of data in different areas inside and outside the mine; Use a multi-level early warning model to identify potential extreme situations. Use a threshold detection algorithm based on historical data and a rule base to determine whether extreme conditions exist. In the event of an extreme condition, send a real-time alert to management personnel and trigger the system's automatic response module. The prediction model of air flow and gas concentration changes in the mine is continuously updated through the adaptive neural network algorithm. When a sharp fluctuation in environmental data is detected, the neural network algorithm immediately calculates the optimal ventilation adjustment strategy and sends instructions to related fan equipment to ensure that harmful gases are quickly discharged. Adopting a reinforcement learning-based algorithm, the system pre-trains ventilation line switching schemes for various extreme situations in a virtual simulation environment. When the main ventilation line is blocked, the system automatically selects the optimal backup ventilation line based on the current working conditions to ensure that the airflow in key areas is not interrupted. By introducing an expert system and combining the knowledge base with a deep learning algorithm, the nonlinear relationship between multiple variables in complex situations can be analyzed. When the system algorithm encounters uncertain situations or data anomalies, the expert system is triggered to compare and analyze historical cases to assist in making key decisions and ensure that the algorithm has redundant safety protection when executing strategies.

2. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 1 is characterized in that: The detailed steps of synchronously processing sensor data and generating a unified environmental perception model through data fusion technology are as follows: First, real-time data is collected from a variety of sensors placed in different areas of the mine; Perform time synchronization and space alignment processing; The data from different types of sensors are fused and processed, and a multi-source information fusion algorithm is used to unify the data from multiple sensors into an environmental perception model; After completing data fusion, the system generates a unified environmental perception model inside and outside the mine and updates it dynamically in real time.

3. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 1 is characterized in that: The detailed steps of the multi-level early warning model to identify potential extreme situations and trigger the automatic response module are as follows: The system first obtains environmental data from sensors in the mine and compares the acquired environmental data with historical data; the multi-level early warning model gradually identifies potential extreme conditions by setting multiple risk levels, and each level corresponds to a different value range and response strategy; Using the threshold detection algorithm based on historical data and rule base, real-time analysis is performed to determine whether the collected environmental data has reached the critical value of extreme working conditions; Once the system detects an extreme working condition, it immediately triggers the preset alarm system and notifies the mine management personnel through multiple channels. It also generates a real-time alarm report based on the mine emergency plan and records the entire event detection and response process through the log function to facilitate subsequent analysis and optimization. When extreme working conditions are detected, the preset emergency module is automatically triggered.

4. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 1 is characterized in that: The detailed steps of updating the prediction model of air flow and gas concentration in the mine and optimizing the ventilation volume based on the adaptive neural network algorithm are as follows: First, the neural network model receives real-time environmental data collected by sensors as input, including gas concentration C(t), temperature T(t), humidity H(t), air flow rate v(t) and oxygen content O2(t), and the input data is composed of a vector X(t) = C(t), T(t), H(t), v(t), O2(t) and input into the adaptive neural network model in real time. The neural network uses a recursive neural network or a long short-term memory network to capture the complex time series relationship in the mine, and uses the following formula to update the internal state of the model: h t =σ(W h ·h t-1 +W x ·X(t)+b h ), where h t is the hidden layer state, W h and W x are the weights of the state transfer matrix and the input matrix, respectively, h is the bias vector, σ is the activation function, and its internal state is gradually updated by continuously inputting time series data, providing a basis for the subsequent prediction and optimization of ventilation volume; Through the training of adaptive neural networks, the changing rules of air flow and gas concentration in the mine are continuously learned, and the prediction model is updated in real time. The prediction model is based on a multi-step prediction strategy to calculate the gas concentration and air flow rate at the future time t+Δt. The prediction process is expressed by the following formula: =f(h t )+∈, in, and are the predicted future gas concentration and air flow rate respectively; f and g are the nonlinear mapping functions of the neural network; ∈ and η are noise terms used to simulate random interference factors in the mine.

5. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 4 is characterized in that: When a sharp fluctuation in the mine environment data is detected, the neural network immediately calculates the optimal ventilation adjustment strategy to achieve rapid discharge of harmful gases. The optimization strategy is solved based on the following objective function: ， Where, Q(t) is the ventilation volume of the fan at time t; C safe is the gas concentration under the safety threshold; v target is the target air flow rate; E(Q) is the energy consumption function of the ventilation system; α, β, γ are weight parameters that control the balance between safety, flow rate and energy consumption.

6. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 1 is characterized in that: The detailed steps of virtual simulation training and optimal ventilation route switching based on reinforcement learning algorithm are as follows: First, a virtual simulation environment is built to simulate the actual terrain structure and ventilation system of the mine, including the layout of the air ducts, the distribution of the main and backup fans, and the key operating areas of the mine; Using a reinforcement learning framework, the mine ventilation system is defined as the "environment" and the algorithm as the "agent". The algorithm learns how to make optimal decisions under different extreme working conditions by constantly interacting with the virtual environment. After a lot of training, the reinforcement learning algorithm forms a policy network to select the optimal ventilation path in different scenarios; When the main ventilation line is blocked, the system immediately calls the optimal backup ventilation plan calculated by the reinforcement learning algorithm and automatically performs the switching operation. The algorithm will quickly select a path to ensure uninterrupted airflow in key areas based on current sensor data and status.

7. The method for intelligently controlling a coal mine ventilation system based on environmental perception according to claim 1 is characterized in that: The specific steps for implementing decision support combining expert system and deep learning algorithm are as follows: Deep learning models are introduced to handle complex nonlinear relationships. Through training with a large amount of mine data, deep learning models can predict the changing trends and potential risks between environmental variables. The framework combines expert systems and deep learning. When the system receives new data, it triggers corresponding expert rules based on the prediction results of deep learning. Continuously monitor the mine's environmental data and trigger the anomaly detection module when data anomalies are detected; After receiving an abnormal situation, the expert system selects the case that best fits the current situation from the knowledge base through the reasoning engine and compares it with the prediction results of the deep learning model; After completing the joint analysis, the system generates a specific emergency decision-making plan and provides decision-making suggestions to mine managers through a visual interface, while indicating potential risks and implementation steps.

Images