Air return corner gas regulation and control method based on intelligent air guide

Abstract

The invention discloses an air return corner gas regulation and control method based on intelligent air guide. The air return corner gas regulation and control method comprises the following steps that S1, an air guide curtain is installed; s2, acquiring a state value and an action value of the intelligent agent, and calculating a reward value based on the state value and the action value; s3, the step S2 is repeated, multiple sets of data are obtained to serve as a training set, the training set is used for training the intelligent agent, and an air guide curtain regulation and control model with the state value as input and the action value as output is obtained; s3, environment information at the air return corner of the current working face is collected, whether the oxygen concentration is smaller than a threshold value or not is judged, if yes, the next step is executed, and if not, environment information continues to be collected; and S4, inputting the environment information and the state information of the air guide curtain into the air guide curtain regulation and control model to obtain a regulation and control action, and returning to S3 after the air guide curtain executes the action. The environment state can be sensed in real time, regulation and control actions are automatically generated and executed when the oxygen concentration is lower than a threshold value, and the form of the air guide curtain is dynamically adjusted, so that the ventilation condition of the air return corner is improved, and the safety level of the underground operation environment is improved.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety technology, specifically a method for regulating return air corner gas based on intelligent ventilation. Background Technology

[0002] During the longwall mining of top coal caving faces, excessive concentrations of gases such as CO and CO2 are observed in the return air corners, leading to low oxygen levels in these areas. This severely impacts mine safety and may even cause shutdowns, significantly reducing mine efficiency. Installing air guide curtains in the return air corners aims to reduce harmful gas concentrations and increase oxygen levels by guiding airflow direction and speed. However, traditional air guide curtains largely rely on manual adjustment, requiring workers to rely on experience to judge airflow direction and corner oxygen concentrations, manually adjusting the angle and position of the curtains. This method is not only labor-intensive but also prone to operational errors and delays, failing to respond promptly to dynamically changing mine environments. Furthermore, existing air guide curtains are typically fixed installations, unable to be flexibly adjusted according to face conditions. When airflow direction changes or mining equipment moves, the curtains cannot be adjusted in time, resulting in poor ventilation. Finally, existing air guide curtains lack integration with sensor networks or data analysis systems, failing to provide real-time data feedback and adjustment suggestions. This means that optimization and adjustments cannot be made based on data, and the ventilation effect relies on traditional experience-based judgment, which lacks scientific rigor and precision. Summary of the Invention

[0003] The purpose of this invention is to provide a method for regulating return air corner gas based on intelligent air guidance, comprising the following steps:

[0004] S1. Install an air guide curtain on the side of the hydraulic support at the return air corner near the return air corner;

[0005] S2. Obtain environmental information at the return air corner of the working face before and after executing the control action, as well as the status information of the air guide curtain, as the state value of the intelligent agent;

[0006] S3. Obtain the control action of the air guide curtain at the previous moment and use it as the action value of the intelligent agent;

[0007] S4. Determine the industry control objectives according to relevant industry standards, and then, based on these objectives, design the reward function for the intelligent agent and calculate the reward value based on the state value and action value.

[0008] S5. Repeat steps S2) to S4) to obtain multiple sets of data as training sets;

[0009] S6. Train the agent using the training set to obtain a reinforcement learning-based air curtain control model with state values ​​as input and action values ​​as output.

[0010] S7. Collect environmental information at the current working face return air corner using sensors;

[0011] Determine if the oxygen concentration is less than the safety threshold. If yes, proceed to step S8; otherwise, continue collecting environmental information at the return air corner of the current working face.

[0012] S8. Input the environmental information and the current state information of the air guide curtain into the reinforcement learning-based air guide curtain control model to obtain the control action of the air guide curtain.

[0013] S9. After the air guide curtain performs its action, return to step S7.

[0014] Furthermore, in step S2), the environmental information includes oxygen concentration, methane concentration, carbon monoxide concentration, wind speed, wind direction, temperature, and humidity.

[0015] Several measuring points are set at the return air corner of the working face. The measuring points are located at the middle position near the top plate, bottom plate, and roadway height, and are arranged longitudinally. The measuring points are equipped with oxygen sensors, methane sensors, carbon monoxide sensors, ultrasonic wind direction and speed sensors, temperature sensors, and humidity sensors to monitor environmental information.

[0016] Furthermore, in step S3), the control action includes increasing / decreasing the opening and closing angle of the air guide curtain blades and increasing / decreasing the extension length of the air guide curtain.

[0017] Furthermore, in step S4), the reward function As shown below:

[0018] (1)

[0019] In the formula: , , These represent the changes in oxygen concentration, methane concentration, and carbon monoxide concentration, respectively. The cost incurred in adjusting the air deflector curtain; , , , These are the weighting functions for oxygen, methane, and carbon monoxide, respectively.

[0020] Furthermore, in step S6), the training method for the reinforcement learning-based air curtain control model includes the following steps:

[0021] S6.1, Set the current state value The input is fed into the policy network to calculate the probability of selecting each action value in the current state and generate a probability distribution;

[0022] S6.2. Based on the probability distribution, randomly select an action value. This is used as an action to perform, and after performing the action, a reward value is calculated based on the reward function. Record reward value and new state values ;

[0023] S6.3, Set the current state value Action value Reward Value New state value Save it to the memory bank as the training set; determine... Is it equal to the maximum number of loops? If yes, proceed to step S6.4; otherwise, let... = +1, and return to step S6.1).

[0024] S6.4. Use the value network to score the state values ​​in the training set, and based on the current state value... New state value and reward value Calculate actions Advantages ;

[0025] Advantage value The calculation formula is as follows:

[0026] (2)

[0027] (3)

[0028] In the formula: In the state Next, select an action. Additional advantages compared to choosing the average action; A state-action function, representing the state... Next action The expected cumulative return that can be obtained afterward; / The state-value function represents the state value. / The expected cumulative return that can be obtained by following the current strategy; Discount factor;

[0029] S6.5 Calculate the probability of choosing the original action under the same conditions using both the old and new policy networks, and calculate the probability ratio between the two.

[0030] S6.6, Constructing the loss function for the policy network And based on probability ratios and advantage values Calculate the loss function ;judge If the value is less than a threshold, output the policy network; otherwise, update the parameters of the policy network using gradient descent and set... = +1, clear the memory bank;

[0031] Loss function of policy network As shown below:

[0032] (4)

[0033] In the formula: Choosing actions for new and old strategies The probability ratio above; This is the clipping threshold;

[0034] S6.7 Loss Function for Constructing a Value Network And based on scores and advantage values Calculate the loss function ;judge If the value is less than a threshold, output the value network; otherwise, update the parameters of the value network using gradient descent and let... = +1, return to step S6.1); loss function of the value network. As shown below:

[0035] (5)

[0036] In the formula: This represents the current predicted value of the value network. For target value.

[0037] Furthermore, in step S6.5), the range of the probability ratio is [1- 1+ ], The pruning threshold is in the range [0.1, 0.3].

[0038] Furthermore, in step 1), several hydraulic supports are installed at the coal mining face, one of which is located at the return air corner and is referred to as the return air corner hydraulic support.

[0039] An air guide curtain is installed on the hydraulic support at the return air corner. The air guide curtain includes an air guide curtain body and a sensor group.

[0040] The air guide curtain body includes a support component and a blade actuator unit. The support component is fixed to a hydraulic bracket at the return air corner and is used to support the movement of the blade actuator unit. The blade actuator unit includes a telescopic drive explosion-proof electric push rod, a rotary drive explosion-proof electric push rod, and several blades. The telescopic drive explosion-proof electric push rod and the rotary drive explosion-proof electric push rod are used to execute control commands from the microprocessor to control the movement of the air guide curtain.

[0041] Each blade is press-fitted with a rotating shaft. When the blade is fully retracted, the outermost blade facing the roadway at this time is called the head blade, and the blade at the other end is called the tail blade.

[0042] The ends of the first and last blades are equipped with sensor groups, which include displacement sensors and rotary encoders, respectively used to monitor the current deployment position and opening / closing angle of the air guide curtain.

[0043] Furthermore, when the blades of the air guide curtain are extended, the extended portion of the air guide blades has an overall arc-shaped structure, with the inner arc surface facing the air intake direction.

[0044] Furthermore, in step 1), a distance sensor is installed on the air guide curtain to monitor the distance between the coal mining equipment and the air guide curtain.

[0045] When the control unit detects that the coal mining equipment is approaching the air curtain, it sends a retraction command to the actuator drive system, which uses a servo motor to retract the air curtain.

[0046] Furthermore, in step 1), a wind vane is installed on the air guide curtain.

[0047] The wind vane is installed on the windward side of the top of the air guide curtain to sense the wind direction in the mine roadway in real time. The wind direction data is used as the state value of the intelligent agent and input into the air guide curtain control model based on reinforcement learning, so that the intelligent agent can dynamically adjust the windward angle of the air guide curtain according to the wind direction change.

[0048] The technical effects of this invention are undeniable, and its beneficial effects are as follows:

[0049] 1. This invention, through its retractable design, enables the air curtain to automatically retract or expand when needed, adapting to different working environments and operational requirements.

[0050] 2. This invention acquires environmental data in real time by installing multiple sensors (oxygen concentration sensor, wind pressure sensor, etc.) at different positions in the return air corner and the air guide curtain. Through multi-sensor fusion technology, it performs precise spatial positioning and analysis of airflow, thereby more accurately controlling the dynamic adjustment of the air guide curtain.

[0051] 3. The air guide curtain in this invention utilizes machine learning algorithms combined with sensor data to automatically adjust its shape and position according to real-time environmental changes, thereby controlling airflow more efficiently, improving oxygen concentration in the return air corner, and enhancing mine safety and production efficiency. This method has advantages such as strong adaptability, rapid response, and continuous optimization, and can maintain good performance in different mine environments. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the low-oxygen "guiding" method of the air guide curtain of the present invention;

[0053] Figure 2 for Figure 1 A magnified view of a section at point A in the middle;

[0054] Figure 3 A diagram illustrating reinforcement learning;

[0055] In the diagram: 1-Return air corner; 2-Return air corner hydraulic support; 3-Air guide curtain. Detailed Implementation

[0056] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0057] Example 1:

[0058] A method for regulating return air corner gas based on intelligent air guidance includes the following steps:

[0059] S1. Install the air guide curtain 3 on the side of the hydraulic support 2 near the return air corner 1;

[0060] S2. Obtain environmental information at the return air corner 1 of the working face before and after the execution of the control action, as well as the status information of the air guide curtain 3, as the status value of the intelligent agent;

[0061] S3. Obtain the control action of the air guide curtain 3 at the previous moment, and use it as the action value of the intelligent agent;

[0062] S4. Determine the industry control objectives according to relevant industry standards, and then, based on these objectives, design the reward function for the intelligent agent and calculate the reward value based on the state value and action value.

[0063] S5. Repeat steps S2) to S4) to obtain multiple sets of data as training sets;

[0064] S6. Train the agent using the training set to obtain a reinforcement learning-based air curtain control model with state values ​​as input and action values ​​as output.

[0065] S7. Collect environmental information at the current working face return air corner 1 using sensors;

[0066] Determine if the oxygen concentration is less than the safety threshold. If yes, proceed to step S8. Otherwise, continue collecting environmental information at the current working face return air corner 1.

[0067] S8. Input the environmental information and the current state information of the air guide curtain 3 into the air guide curtain control model based on reinforcement learning to obtain the control action of the air guide curtain.

[0068] S9. After the air guide curtain performs its action, return to step S7.

[0069] Example 2:

[0070] The main structure of this embodiment is the same as that of Embodiment 1. Further, in step S2), the environmental information includes oxygen concentration, methane concentration, carbon monoxide concentration, wind speed, wind direction, temperature, and humidity.

[0071] Several measuring points are set at the return air corner 1 of the working face. The measuring points are located at the middle position near the top plate, bottom plate, and roadway height, and are arranged longitudinally. The measuring points are equipped with oxygen sensors, methane sensors, carbon monoxide sensors, ultrasonic wind direction and speed sensors, temperature sensors, and humidity sensors to monitor environmental information.

[0072] Example 3:

[0073] The main structure of this embodiment is the same as any one of embodiments 1 to 2. Further, in step S3), the control action includes increasing / decreasing the opening and closing angle of the blades in the air guide curtain 3 and increasing / decreasing the extension length of the air guide curtain 3.

[0074] Example 4:

[0075] The main structure of this embodiment is the same as any one of embodiments 1 to 3. Further, in step S4), the reward function... As shown below:

[0076] (1)

[0077] In the formula: , , These represent the changes in oxygen concentration, methane concentration, and carbon monoxide concentration, respectively. The cost incurred in adjusting the air deflector curtain; , , , These are the weighting functions for oxygen, methane, and carbon monoxide, respectively.

[0078] Example 5:

[0079] The main structure of this embodiment is the same as any one of embodiments 1 to 4. Further, in step S6), the training method for the reinforcement learning-based air curtain control model includes the following steps:

[0080] S6.1, Set the current state value The input is fed into the policy network to calculate the probability of selecting each action value in the current state and generate a probability distribution;

[0081] S6.2. Based on the probability distribution, randomly select an action value. This is used as an action to perform, and after performing the action, a reward value is calculated based on the reward function. Record reward value and new state values ;

[0082] S6.3, Set the current state value Action value Reward Value New state value Save it to the memory bank as the training set; determine... Is it equal to the maximum number of loops? If yes, proceed to step S6.4; otherwise, let... = +1, and return to step S6.1).

[0083] S6.4. Use the value network to score the state values ​​in the training set, and based on the current state value... New state value and reward value Calculate actions Advantages ;

[0084] Advantage value The calculation formula is as follows:

[0085] (2)

[0086] (3)

[0087] In the formula: In the state Next, select an action. Additional advantages compared to choosing the average action; A state-action function, representing the state... Next action The expected cumulative return that can be obtained afterward; / The state-value function represents the state value. / The expected cumulative return that can be obtained by following the current strategy; The discount factor is used to balance the weights of immediate rewards and future long-term returns; Equation (3) is the time-series difference estimation method, which uses observed rewards... and future state values Approximate substitute for reality Values ​​that adapt to the real-time requirements of online reinforcement learning training;

[0088] S6.5 Calculate the probability of choosing the original action under the same conditions using both the old and new policy networks, and calculate the probability ratio between the two.

[0089] S6.6, Constructing the loss function for the policy network And based on probability ratios and advantage values Calculate the loss function ;judge If the value is less than a threshold, output the policy network; otherwise, update the parameters of the policy network using gradient descent and set... = +1, clear the memory bank;

[0090] Loss function of policy network As shown below:

[0091] (4)

[0092] In the formula: Choosing actions for new and old strategies The probability ratio above; This is the clipping threshold;

[0093] S6.7 Loss Function for Constructing a Value Network And based on scores and advantage values Calculate the loss function ;judge If the value is less than a threshold, output the value network; otherwise, update the parameters of the value network using gradient descent and let... = +1, return to step S6.1); loss function of the value network. As shown below:

[0094] (5)

[0095] In the formula: This represents the current predicted value of the value network. For target value.

[0096] Example 6:

[0097] The main structure of this embodiment is the same as that of embodiment 5. Further, in step S6.5), the range of the probability ratio is [1- 1+ ], The pruning threshold is in the range [0.1, 0.3].

[0098] Example 7:

[0099] The main structure of this embodiment is the same as any one of embodiments 1 to 6. Further, in step 1), several hydraulic supports are arranged at the coal mining face, one of which is set at the return air corner 1 and is referred to as the return air corner hydraulic support 2.

[0100] An air guide curtain 3 is installed on the return air corner hydraulic support 2. The air guide curtain 3 includes an air guide curtain body and a sensor group.

[0101] The air guide curtain body includes a support component and a blade actuator unit. The support component is fixed to the return air corner hydraulic bracket 2 and is used to support the movement of the blade actuator unit. The blade actuator unit includes a telescopic drive explosion-proof electric push rod, a rotary drive explosion-proof electric push rod, and several blades. The telescopic drive explosion-proof electric push rod and the rotary drive explosion-proof electric push rod are used to execute the control commands of the microprocessor to control the movement of the air guide curtain.

[0102] Each blade is press-fitted with a rotating shaft. When the blade is fully retracted, the outermost blade facing the roadway at this time is called the head blade, and the blade at the other end is called the tail blade.

[0103] The ends of the first and last blades are equipped with sensor groups, which include displacement sensors and rotary encoders, respectively used to monitor the current deployment position and opening / closing angle of the air guide curtain.

[0104] Example 8:

[0105] The main structure of this embodiment is the same as that of embodiment 7. Furthermore, when the blades of the air guide curtain 3 are extended, the extended portion of the air guide blades has an overall arc-shaped structure, with the inner arc surface facing the air inlet direction.

[0106] Example 9:

[0107] The main structure of this embodiment is the same as any one of embodiments 1 to 8. Further, in step 1, a distance sensor is installed on the air guide curtain 3. The distance sensor is set on the side rail or the head blade and is used to monitor the distance between the coal mining equipment and the air guide curtain 3.

[0108] When the coal mining equipment is detected to be approaching the air curtain 3, the control unit sends a retraction command to the actuator drive system, and the air curtain 3 is retracted by the servo motor. This can effectively prevent the coal mining equipment from being scratched, pulled or damaged during movement or operation, and ensure the safety of the equipment itself.

[0109] Example 10:

[0110] The main structure of this embodiment is the same as any one of embodiments 1 to 9. Further, in step 1), a wind vane is installed on the air guide curtain 3.

[0111] The wind vane is installed on the windward side of the top of the air guide curtain 3 to sense the wind direction in the mine roadway in real time. The wind direction data is used as the state value of the intelligent agent and input into the air guide curtain control model based on reinforcement learning, so that the intelligent agent can dynamically adjust the windward angle of the air guide curtain 3 according to the wind direction change.

[0112] Example 11:

[0113] The main structure of this embodiment is the same as any one of embodiments 1 to 10. Further, refer to... Figure 1 The specific implementation method is as follows:

[0114] (1) A guide curtain is installed at the return air corner of the fully mechanized mining face to dilute harmful gases in the upper corner and improve ventilation in the upper corner of the working face. The guide curtain includes: guide blades, sensors, and a control unit. The guide blades are adjustable blades, and the airflow direction can be changed by adjusting the angle. The sensors include, but are not limited to, oxygen sensors, temperature sensors, and humidity sensors, for real-time data acquisition. The control unit includes a microprocessor and control circuits, for receiving sensor data and controlling the movement of the guide blades.

[0115] (2) The control unit mentioned in step (1) has a sensor installed on the top of the air guide curtain to monitor the current unfolding state, position and shape of the curtain. The real-time monitoring data of the sensor is fed back to the control module, and the control module analyzes the data to determine whether the air guide curtain needs to be retracted or unfolded.

[0116] Furthermore, the extension and retraction of the air curtain is achieved through an actuator drive system (servo motor). When the sensor detects that the coal mining machine is approaching or the oxygen concentration in the corner reaches the threshold, the actuator drive system will automatically retract or expand the air curtain according to the instructions of the control module, thus avoiding unnecessary interference or damage.

[0117] Meanwhile, the air guide curtain is equipped with an automatic retraction function, which allows the air guide curtain to automatically retract via a mechanical spring roller. When the sensor detects the approach of the coal mining equipment, the motor or spring roller quickly retracts the air guide curtain, and after coal mining is completed, it automatically unfolds back to its original position.

[0118] (3) In step (1), the air guide curtain is designed in an arc shape. This shape can better guide the airflow direction and reduce the energy loss of the airflow during the air guiding process. The arc design can better utilize the changes in wind pressure to form a larger wind pressure difference, making it easier for the airflow to flow to the return wind corner.

[0119] Furthermore, adjustable air deflectors are installed around the air guide curtain. These deflectors dynamically adjust based on changes in corner airflow, direction, and oxygen concentration. The deflectors and air guide curtain work together to more precisely control airflow direction and speed, ensuring effective guidance of airflow to the return air corner.

[0120] (4) The air guide curtain mentioned in step (1) obtains parameters such as oxygen concentration, temperature, and humidity through sensors, and adjusts the opening and closing angle, wind speed, and direction of the air guide curtain through intelligent algorithms, thereby achieving the "guidance" of the low-oxygen environment.

[0121] Furthermore, the intelligent adjustment is characterized by dynamically adjusting the position and shape of the air guide curtain through reinforcement learning. By constructing an agent, the agent continuously explores and learns the optimal strategy in the environment.

[0122] Specifically, it is described as follows: First, define a model for the interaction between an agent and its environment, such as... Figure 2 As shown, in this model: State refers to the current state of the environment, composed of data collected by sensors, including but not limited to oxygen concentration, harmful gas concentration (such as methane, carbon monoxide), wind speed, wind direction, temperature, and humidity. Action refers to the actions the agent can take, i.e., the adjustment operation of the air curtain, including: adjusting the rotation angle of the air curtain (increasing or decreasing the angle), and opening or closing the air curtain (adjusting the curtain length). Reward refers to the feedback the agent receives after taking an action, used to measure the quality of that action. For example:

[0123] a. If adjusting the air deflector curtain increases the oxygen concentration, a positive reward will be given.

[0124] b. If the concentration of toxic gas decreases after adjustment, a positive reward will also be given.

[0125] c. If the adjustment results in a decrease in oxygen concentration or an increase in the concentration of harmful gases, a negative reward will be given.

[0126] A policy is the strategy by which an agent chooses the best action based on the current state. A policy can be a probability distribution or a decision rule that guides the action the agent should take in each state.

[0127] ① Sensors are deployed at different locations in the mine to monitor oxygen concentration, wind speed, wind direction, temperature, humidity, and harmful gas concentrations in real time. This data is then transmitted to edge computing devices or a central control system to provide input for the RL algorithm.

[0128] ② Set the initial position, angle, and state of the air guide curtain, and ensure that it can be adjusted by the actuator.

[0129] ③ A well-designed reward function is crucial for the success of RL algorithms. For the task of adjusting the air curtain, the reward function can be designed as follows:

[0130]

[0131] in, This represents the change in oxygen concentration (positive if it increases).

[0132] This represents the change in methane concentration (negative if it decreases).

[0133] This represents the change in carbon monoxide concentration (negative if it decreases).

[0134] E represents the cost of adjusting the air guide curtain;

[0135] , , , This is a weighting function used to measure the importance of each factor.

[0136] ④ The intelligent agent uses a reinforcement learning model on the computing device to make decisions based on real-time sensor data. Specifically, it adjusts the angle and opening / closing state of the air curtain in real time. For example, when the oxygen concentration is lower than a set threshold (according to coal mine safety production standards, the oxygen concentration threshold σ should not be less than 18%), the intelligent agent immediately calculates the optimal air curtain adjustment operation and executes it.

[0137] Optimization is performed using a hybrid neural network architecture that includes a policy network (Actor) and a value network (Critic) to find the optimal action.

[0138] The policy network parameters are updated by maximizing a pruned objective function. This is achieved by calculating the probability ratio of the new policy to the old policy in selecting actions, and strictly limiting this probability ratio within the interval [1-ε, 1+ε] (ε is the pruning threshold, which controls the magnitude of the change between the new and old policies) to ensure the stability of the process.

[0139] The policy network is evaluated by calculating the dominance function in the value network. The dominance function A(s,a) measures the superiority of taking a specific action a (e.g., increasing the angle of the air deflector by 10°) in state s compared to the average action level following the current policy. A(s,a)>0 indicates that action a should be encouraged; A(s,a)<0 indicates that action a is worse than the average level and should be avoided.

[0140] ⑤ Further, by continuously monitoring sensor data, the effect of the adjustment is judged. If the oxygen concentration and harmful gas concentration improve after the adjustment, the adjustment is maintained or fine-tuned; if the effect is not good, a new adjustment is made according to the RL strategy.

[0141] Example 12:

[0142] The main structure of this embodiment is the same as any one of embodiments 1 to 11. Further, a device for regulating return air corner gas based on intelligent air guidance includes an air guide curtain 3, an oxygen sensor, a methane sensor, a carbon monoxide sensor, an ultrasonic wind direction and speed sensor, a temperature sensor, a humidity sensor, and a microprocessor. The device and its components comply with mining explosion-proof standards and adopt intrinsically safe or explosion-proof structures.

[0143] Several hydraulic supports are installed at the coal mining face, one of which is located at the return air corner 1 and is referred to as the return air corner hydraulic support 2.

[0144] An air guide curtain 3 is installed on the return air corner hydraulic support 2. In this embodiment, the air guide curtain adopts a mine-use intrinsically safe air guiding device with telescopic and rotational functions. The device includes an air guide curtain body and a sensor group. Its core feature is that the air guide curtain body can both extend and retract as a whole along the roadway direction to adjust the effective air guiding distance, and can also deflect at the same angle to accurately guide the airflow direction.

[0145] The air guide curtain body includes a support component and a blade actuator.

[0146] The support components include a side rail, a support frame, a linear main rail, a slider assembly, a bearing seat, an auxiliary guide rail, an auxiliary slider assembly, and an X-shaped linkage structure.

[0147] The side rail is fixed to the return air corner hydraulic support 2. The support frame is a cubic frame, with one end fixed inside the side rail and the other end extending out of the side rail, located in the roadway. The support frame has two crossbeams, I and II, on its frame surface facing away from the return air roadway. Crossbeam I is located above crossbeam II. The support frame has two crossbeams, III and IV, on its frame surface facing the return air roadway. Crossbeam III is located above crossbeam IV. Crossbeams III and IV are spatially staggered with crossbeams I and II, and are respectively located below crossbeams I and II.

[0148] Linear main guide rails are installed on the side walls of crossbeams I and II. Several slider assemblies are spaced apart within the linear main guide rails, and a bearing seat is fixedly installed on the slider body of each slider assembly. The bearing seats on crossbeams I and II are respectively referred to as the upper bearing seat and the lower bearing seat.

[0149] The top surfaces of crossbeams III and IV are equipped with auxiliary guide rails, and several auxiliary slider assemblies are spaced apart within the auxiliary guide rails. Each slider body has two ear plates at its bottom left and right ends, and adjacent slider bodies are connected by an X-shaped linkage structure. The X-shaped linkage structure includes two connecting rods. The two connecting rods are hinged at their middle sections, and the hinge point is connected to the auxiliary slider assembly. The two ends of the X-shaped connecting rod are respectively hinged to the ear plates of the adjacent slider bodies.

[0150] The number of auxiliary slider components is one less than the number of slider components.

[0151] The blade actuator includes a blade, a rotating shaft, a telescopic explosion-proof electric push rod, a rotary explosion-proof electric push rod, a floating support mechanism, a telescopic active linkage rod, a rotary explosion-proof electric push rod, a rocker arm, and a linkage rod.

[0152] The number of blades is the same as the number of slider assemblies. With the blades fully retracted, the outermost blade facing the aisle is designated as the head blade, and the blade at the other end is designated as the tail blade. The corresponding slider assemblies are designated as the head slider and the tail slider. The tail slider is fixed to the linear master rail (but does not slide on the linear master rail).

[0153] A through hole is pre-drilled inside each blade along its length. The rotating shaft passes through the through hole and is press-fitted to the blade. Both ends of the rotating shaft pass through the bearing holes of the upper and lower bearing housings, respectively.

[0154] Two telescopic explosion-proof electric actuators are fixed to the support frame and located at the end away from the side rail. One telescopic explosion-proof electric actuator is located above and controls the first slider on crossbeam I. The other telescopic explosion-proof electric actuator is located below and controls the first slider on crossbeam II. The telescopic ends of the telescopic explosion-proof electric actuators are fixed to the first slider. In operation, driving the telescopic explosion-proof electric actuators pushes / pulls the first slider to move. Through the X-shaped linkage structure, all slider assemblies move synchronously at equal intervals, thereby causing the entire blade array to unfold or retract.

[0155] A floating support mechanism is provided oppositely on the top and bottom of the support frame. The floating support mechanism includes a recessed seat and multiple sets of rollers. Two recessed seats are fixed opposite to each other on the top and bottom of the support frame. Multiple sets of rollers are fixed at intervals inside the recessed seats. Each roller set includes two rollers arranged opposite each other. The retractable active linkage rod is disposed in the recessed seat and located between the two rollers of the roller set, and the rollers provide flexible restraint for the retractable active linkage rod.

[0156] The end of the retractable active linkage rod is connected to the telescopic end of the rotary-driven explosion-proof electric actuator. Two rotary-driven explosion-proof electric actuators are fixed to the top and bottom of the support frame, located at the end closest to the side rail. The rotary-driven explosion-proof electric actuator at the top of the support frame controls the upper retractable active linkage rod. The telescopic-driven explosion-proof electric actuator at the bottom of the support frame controls the lower retractable active linkage rod.

[0157] The retractable active linkage rod comprises multiple rigid short rods, with adjacent short rods connected end-to-end via a universal joint. One end of the rotating shaft, passing through the upper and lower bearing seats, is connected to the rocker arm key. The axis of the rotating shaft is perpendicular to the axis of the rocker arm. The other end of the rocker arm is connected to the short rods of the retractable active linkage rod via a connecting rod.

[0158] The rocker arm has two connecting lugs I at the end furthest from the rotation axis. The short rod has two connecting lugs II. Two bearings are fixed to both ends of the connecting rod, one located between the two connecting lugs I (denoted as bearing I) and the other located between the two connecting lugs II (denoted as bearing II). A pin passes through a hole on one lug I / lug II, the inner hole of bearing I / bearing II, and a hole on the other lug I / lug II, and is locked in place by a cotter pin.

[0159] The ends of the first and last blades are equipped with sensor groups electrically connected to the microprocessor. The sensor groups include a displacement sensor and a rotary encoder. The displacement sensor is fixed to a mounting base at the blade end and moves synchronously with the blade. The rotary encoder is fixed to a rotating shaft.

[0160] In this embodiment, the displacement sensor is the GUD240 mining digital display sensor manufactured by China Coal Technology & Engineering Group Co., Ltd. The rotary encoder is the GBH60 intrinsically safe mining encoder manufactured by Shandong Kuang'an Heavy Industry Co., Ltd.

[0161] Several measuring points are set at the return air corner 1 of the working face, arranged longitudinally and located near the top and bottom of the roadway, and at the midpoint of the roadway height. Each measuring point is equipped with an oxygen sensor, a methane sensor, a carbon monoxide sensor, an ultrasonic wind direction and speed sensor, a temperature sensor, and a humidity sensor. By simultaneously collecting full-section environmental data at the longitudinally distributed measuring points in the roadway top, bottom, and middle, the microprocessor can comprehensively acquire gas concentration, airflow parameters, and temperature and humidity information at different heights of the return air corner.

[0162] In this embodiment, the oxygen sensor is the GYH25 intrinsically safe mining sensor manufactured by Libaolai Technology Co., Ltd. The methane sensor is the KGJ28A intrinsically safe sensor manufactured by the Changzhou Research Institute of China Coal Technology & Engineering Group. The carbon monoxide sensor is the H4CO high-end electrochemical sensor manufactured by Shanghai Songbai Sensor Technology Co., Ltd. The ultrasonic wind direction and speed sensor is the TH-WFB5 integrated mining sensor manufactured by a professional explosion-proof meteorological sensor company. The temperature sensor is the WZP2-231 explosion-proof mining sensor manufactured by Anhui Tiankang Group Co., Ltd. The humidity sensor is the KSH05 intrinsically safe sensor manufactured by Jiangsu Sanheng Technology Co., Ltd.

[0163] The device also includes a rotating component; the rotating component includes a top rotating shaft 4, a steering drive servo motor, a worm gear reducer, and an angle feedback sensor. The air guide curtain 3 and the return air corner hydraulic support 2 are connected via the top rotating shaft 4. The steering drive servo motor is electrically connected to a microprocessor. The motor shaft of the steering drive servo motor is connected to the top rotating shaft via a worm gear reducer. The angle feedback sensor is fixed to the end of the top rotating shaft 4 and is electrically connected to the microprocessor.

[0164] When the oxygen concentration in the return air corner is below 18%, methane / carbon monoxide exceeds the standard, or the airflow is turbulent, the microprocessor combines sensor data and reinforcement learning model to dynamically determine the blade extension length (maximum not exceeding 2.0m) and opening angle according to the degree of exceedance, and issues control commands to drive the servo motor and micro geared motor to adjust the air guide curtain 3.

Claims

1. A method for regulating air return corner gas based on intelligent air deflection, characterized in that, The method comprises the following steps: S1. A wind guide curtain (3) is installed on one side of the return air corner hydraulic support (2) close to the return air corner (1); S2. The environmental information at the return air corner (1) of the working face before and after the execution of the control action and the state information of the wind guide curtain (3) are obtained as the state value of the agent; S3. The control action of the wind guide curtain (3) at the previous time is obtained as the action value of the agent; S4. The industry control target is determined according to the relevant industry specifications, and then the agent design reward function is confirmed based on the target, and the reward value is calculated based on the state value and the action value; S5. Steps S2) to S4) are repeated to obtain multiple sets of data as a training set; S6. The training set is used to train the agent to obtain a wind guide curtain control model based on reinforcement learning with the state value as the input and the action value as the output; S7. The environmental information at the return air corner (1) of the current working face is collected by a sensor; It is judged whether the oxygen concentration is less than a safety threshold value, if yes, step S8) is entered, otherwise, the environmental information at the return air corner (1) of the current working face is continuously collected; S8. The environmental information and the state information of the current wind guide curtain (3) are input into the wind guide curtain control model based on reinforcement learning to obtain the control action of the wind guide curtain; S9. After the wind guide curtain executes the action, step S7) is returned.

2. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step S2), the environmental information includes oxygen concentration, methane concentration, carbon monoxide concentration, wind speed, wind direction, temperature and humidity; A plurality of measuring points are arranged at the return air corner (1) of the working face, and the measuring points are arranged in a longitudinal direction at positions close to the roof, the floor and the middle of the roadway height, and oxygen sensors, methane sensors, carbon monoxide sensors, ultrasonic wind direction and speed sensors, temperature sensors and humidity sensors are arranged on the measuring points to monitor the environmental information.

3. The method according to claim 1, wherein the method is characterized by: In step S3), the control action includes increasing / decreasing the opening angle of the blades of the wind guide curtain (3) and increasing / decreasing the extension length of the wind guide curtain (3).

4. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step S4), the reward function As follows: （1） In the formula: , , respectively are the change amount of oxygen concentration, methane concentration, carbon monoxide concentration; is the cost consumed by adjusting the air guide curtain; , , , respectively are the weight functions of oxygen, methane, and carbon monoxide.

5. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step S6), the training method of the wind guide curtain control model based on reinforcement learning comprises the following steps: S6.1, input the current state value into the policy network, calculate the selection probability of each action value under the current state, and generate a probability distribution; S6.2, randomly select an action value based on the probability distribution , as an execution action, and after executing the action, calculate a reward value based on the reward function , record the reward value and the new state value ; S6.3, save the current state value , action value , reward value , new state value to the memory bank as a training set; determine whether it is equal to the maximum number of loops, if so, go to step S6.4), otherwise, let = + 1, and return to step S6.1); S6.4, score the state values in the training set using the value network and based on the current state value , the new state value , and the reward value , compute the advantage value for the action ; Advantage value The calculation formula is as follows: （2） （3） where: is the state , select action with respect to the additional advantage of selecting the average action; is the state-action function representing the expected cumulative reward obtainable in state by performing action ; / is the state value function representing the expected cumulative reward obtainable in state by following the current policy; / is the discount factor; S6.

5. The probabilities of selecting the original action under the same state are calculated using the new and old policy networks respectively, and the probability ratio of the two is calculated; S6.6, loss function of constructing strategy network , and based on probability ratio and advantage value , calculate loss function ; determine whether is less than threshold value, if yes, output strategy network, otherwise, update parameters of strategy network based on gradient descent method, and let = +1, clear memory bank; Loss function for policy network As follows: （4） wherein: is the ratio of the probabilities of the old and new policies on the action chosen. is the clipping threshold. S6.7 Loss Function for Constructing a Value Network And based on scores and advantage values Calculate the loss function ;judge If the value is less than a threshold, output the value network; otherwise, update the parameters of the value network using gradient descent and let... = +1, return to step S6.1). Loss function of value network As follows: （5） In the formula: is the current predicted value of the value network; is the target value.

6. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step S6.5), the probability ratio ranges from [1- , 1+ ], is a cropping threshold ranging from [0.1, 0.3].

7. The method of claim 1, wherein the method is based on intelligent air deflection. In step 1), a plurality of hydraulic supports are arranged at the position of the coal mining face, one of which is arranged at the return air corner (1) and is denoted as the return air corner hydraulic support (2); The wind guide curtain (3) is installed on the hydraulic support (2); the wind guide curtain (3) comprises a wind guide curtain body and a sensor group; The wind guide curtain body comprises a support component and a blade execution unit; the support component is fixed on the return air corner hydraulic support (2) and is used to support the movement of the blade execution unit; the blade execution unit comprises a telescopic explosion-proof electric push rod, a rotary explosion-proof electric push rod and a plurality of blades; the telescopic explosion-proof electric push rod and the rotary explosion-proof electric push rod are used to execute the control instructions of the microprocessor to control the movement of the wind guide curtain; A rotating shaft is arranged on each blade; when the blades are fully retracted, the outermost blade facing the roadway at this time is denoted as the head blade, and the blade at the other end is denoted as the tail blade; The end of the head blade and the tail blade is provided with a sensor group, which includes a displacement sensor and a rotary encoder, respectively used for monitoring the current unfolding position and the opening and closing angle of the air guide curtain.

8. The air return corner gas regulation method based on intelligent air deflector according to claim 7, characterized in that: In the extended state of the air guide curtain (3), the air guide blades of the extended part are in an arc structure as a whole, and the inner arc surface faces the air inlet direction.

9. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step 1), the distance sensor is installed on the air guide curtain (3) to monitor the distance between the coal mining equipment and the air guide curtain (3); When it is monitored that the coal mining equipment approaches the air guide curtain (3), the control unit issues a retracting instruction to the actuator driving system, and the servo motor is used to retract the air guide curtain (3).

10. The air return corner gas regulation method based on intelligent air deflector according to claim 1, characterized in that: In step 1), the air guide curtain (3) is provided with an air direction plate; The air direction plate is installed on the windward side of the top of the air guide curtain (3) to sense the wind direction in the mine tunnel in real time, and the wind direction data is input as the state value of the intelligent agent into the air guide curtain control model based on reinforcement learning, so that the intelligent agent dynamically adjusts the windward angle of the air guide curtain (3) according to the change of the wind direction.

Images