A method and system for joint optimization and control of intelligent ventilation in mines

Abstract

The present invention discloses a method and system for joint optimization and control of intelligent mine ventilation, including: S1: collecting multi-source mine ventilation data through distributed multimodal acoustic wave sensors using a grid deployment strategy; S2: using trust function theory to fuse and optimize the multi-source mine ventilation data to obtain comprehensive state data; S3: based on the comprehensive state data, obtaining the optimal preliminary control action of different mine equipment through an adaptive reinforcement learning model; S4: based on the optimal preliminary control action, introducing the tunnel structure safety as a dynamic constraint factor and combining with the improved pigeon flock optimization algorithm to obtain the final control action, and realizing underground ventilation control based on the final control action; by incorporating the tunnel structure safety into the fitness function, the algorithm fully considers the safety of the tunnel structure while optimizing the ventilation effect and energy consumption.

Description

Technical Field

[0001] The present invention relates to the field of mine control technology, and in particular to a method and system for combined optimization and control of intelligent mine ventilation. Background Art

[0002] In mine mining operations, the ventilation system plays a key role in ensuring safe production underground. Mine ventilation must not only ensure that the underground air quality meets the standards, effectively dilute and discharge harmful gases such as gas, but also maintain a suitable working environment temperature and humidity. As the depth and breadth of mine mining increase, the ventilation system faces more challenges. On the one hand, the underground ventilation network structure is becoming more and more complex, and the ventilation needs in different areas vary significantly, and are constantly changing due to factors such as geological conditions and mining progress. On the other hand, the operating status of ventilation equipment will also change over time. Equipment failure may cause ventilation abnormalities. If they cannot be detected and handled in a timely and accurate manner, they may easily lead to safety accidents.

[0003] At present, the existing ventilation control methods are insufficient in ensuring the safety of tunnel structures. The changes in wind pressure and wind speed caused by ventilation may affect the stability of the tunnel and threaten the safety of underground workers and equipment. In addition, the existing mine ventilation control technology is difficult to comprehensively consider multiple factors such as ventilation effect, energy consumption and tunnel structure safety. Therefore, a method and system for joint optimization and control of intelligent mine ventilation is proposed here. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention proposes the following technical solutions:

[0005] A method and system for joint optimization and control of intelligent mine ventilation, comprising:

[0006] A method and system for joint optimization and control of intelligent ventilation in mines, the method steps are as follows:

[0007] S1: Collect multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy;

[0008] S2: Use trust function theory to integrate and optimize multi-source mine ventilation data to obtain comprehensive status data;

[0009] S3: Based on the comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal preliminary control actions for different mine equipment;

[0010] S4: Based on the optimal preliminary control action, the tunnel structure safety factor is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is achieved based on the final control action;

[0011] Among them, the improved pigeon flock optimization algorithm is implemented by using chaotic mapping to initialize the pigeon flock position in the pigeon flock optimization algorithm, introducing a diversity feedback mechanism in the pigeon flock position update process, and considering the safety of the tunnel structure.

[0012] The process of collecting multi-source mine ventilation data is as follows:

[0013] Deploy multimodal acoustic wave sensors and space adjacent sensor nodes horizontally at a distance of , the vertical spacing distance is , forming a spatial grid, each sensor node is equipped with a low-power acoustic wave transmitting and receiving module, and each sensor node is set according to the sampling frequency Acoustic wave signals in the mine are collected. The collected acoustic wave signals are discrete signal sequences, which include wind speed data, tunnel structure acoustic signals and equipment speed signals.

[0014] The process of fusion optimization of multi-source mine ventilation data using trust function theory is as follows:

[0015] Define an identification framework, which includes "ventilation equipment normal" and "ventilation equipment failure";

[0016] Feature extraction is performed on multi-source mine ventilation data collected from a distributed acoustic sensor array. The time domain signal is converted into a frequency domain signal through fast Fourier transform, and the spectral characteristics of the signal are extracted. Based on the extracted spectral characteristics, an evidence body is constructed. The evidence body includes both normal ventilation equipment and ventilation equipment failure.

[0017] Based on the spectrum characteristics, the basic probability distribution function is determined by comparing them with the pre-set fault characteristic threshold. For the spectrum characteristic data of all sensor nodes, each node constructs an evidence body based on its own collected data and uses the Dempster-Shafer synthesis rule to obtain the trust distribution data.

[0018] Based on the trust distribution data, the trust distribution data of each proposition is arranged in sequence according to the order of the propositions in the proposition set to form a data sequence. This data sequence is the comprehensive status data.

[0019] The process of obtaining trust distribution data using the Dempster-Shafer synthesis rule is as follows:

[0020] Based on the evidence bodies of any two sensor nodes, the evidence bodies are fused to obtain the fused basic probability distribution. The evidence bodies of all sensor nodes are traversed to perform Dempster-Shafer synthesis rule fusion to obtain the corresponding basic probability distribution. The basic probability distributions of all sensor nodes are combined together to obtain the trust distribution data.

[0021] The adaptive reinforcement learning model uses a deep neural network as its basic architecture, introduces the local perception and weight sharing ideas of convolutional neural networks in the hidden layer, and performs strategy updates based on a dual-deep Q-network algorithm to adapt to the complex nonlinear dynamic characteristics of the mine ventilation system.

[0022] The adaptive reinforcement learning model includes an input layer, multiple hidden layers and an output layer.

[0023] The initial control action acquisition process is as follows:

[0024] The input layer of the adaptive reinforcement learning model receives the comprehensive state data, converts it into a comprehensive vector representing the dimension, and then transmits it to the hidden layer;

[0025] The hidden layer consists of multiple neurons, which are connected by weights. The activation function is used to perform nonlinear transformation on the output of the neurons. The local perception and weight sharing ideas of convolutional neural networks are introduced. Convolution modules are set at intervals to optimize the convolution operation on local areas of the data.

[0026] The output layer outputs the control action scores of different devices. The number of neurons corresponds to the number of device actions controlled, and the output layer outputs the probability of control actions of different devices.

[0027] Strategy update is performed through the dual-depth Q network algorithm. The dual-depth Q network includes the current network and the target network. The Q function is defined. The Q function includes the parameters of the deep neural network. The current network is used to select actions, and the target network evaluates the value of actions. The current network parameters are updated by minimizing the loss function and adopting the experience revisit mechanism. After multiple rounds of training, the current network parameters are updated. When the training converges, the optimal preliminary control strategy under different comprehensive states is obtained.

[0028] The process of obtaining the safety degree of the tunnel structure is as follows:

[0029] Using safety sensors to monitor the deformation and stress parameters of the tunnel, the safety sensors include displacement sensors and stress sensors;

[0030] The deformation and stress parameters are combined with the geomechanical model to obtain the tunnel structure safety index, and the tunnel structure safety degree is quantified based on the structural safety index through the empirical model.

[0031] The process of obtaining the final control action by introducing the tunnel structure safety as a dynamic constraint factor and combining it with the improved pigeon flock optimization algorithm is as follows:

[0032] Introducing the pigeon flock optimization algorithm and using chaos mapping to initialize the pigeon flock position

[0033] Assume the size of the pigeon flock is , each pigeon represents a combination of control actions, the vector dimension is consistent with the number of parameters of the control action, and the position of the initial pigeon group within the parameter value range is randomly generated;

[0034] A fitness function is used to evaluate the pros and cons of each control action combination. For each control action combination, its fitness value is calculated according to the fitness function, and the position of the control action combination with the highest fitness is found, which is the global optimal position.

[0035] Adjust the flight direction of the pigeon flock according to the global optimal position, obtain the flight speed, and update the position according to the flight speed;

[0036] A diverse feedback mechanism is introduced during the position update process. Based on the tunnel structure safety and the correction amount determined by the control action adjustment rules, when the pigeon's position causes the tunnel structure safety to fall below the preset safety threshold, its position is corrected using the correction amount.

[0037] Repeatedly calculate the fitness function, update the position and speed of the pigeon group until the iteration termination condition is met, the fitness function converges or the maximum number of iterations is reached, and the global optimal position is finally obtained. The corresponding control action combination is the final control action.

[0038] The process of determining the correction amount according to the tunnel structure safety and the control action adjustment rules is as follows:

[0039] The tunnel structure safety index is obtained based on the tunnel structure safety threshold and the actual tunnel structure safety corresponding to the pigeon position;

[0040] Construct a control action adjustment rule, which includes the fan speed and ventilation valve opening. Set the fan speed correction value to be and the ventilation valve opening correction value to be. The fan speed correction value is obtained based on the minimum step length of the fan speed adjustment considered in the current control action combination. The ventilation valve opening correction value is obtained based on the minimum step length of the ventilation valve opening adjustment considered in the current control action combination:

[0041] The final correction amount is determined based on the fan speed correction amount and the ventilation valve opening correction amount.

[0042] The implementation process of the diversity feedback mechanism is as follows:

[0043] Calculate the diversity index of the pigeon flock and use the Euclidean distance to measure the average distance between pigeons. When the diversity index is lower than the preset threshold, randomly perturb the position of the pigeons.

[0044] The diversity index is obtained by calculating the Euclidean distance between any two pigeons, adding up the Euclidean distances between all pairs of pigeons, and dividing the sum by the total number of pairwise combinations of pigeons.

[0045] A mine intelligent ventilation joint optimization and control system, comprising:

[0046] Data acquisition module: collects multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy;

[0047] Fusion optimization module: Use trust function theory to fuse and optimize multi-source mine ventilation data to obtain comprehensive status data;

[0048] Reinforcement Learning Module: Based on comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal initial control actions for different mine equipment;

[0049] Control optimization module: Based on the optimal preliminary control action, the tunnel structure safety is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is realized based on the final control action.

[0050] The present invention has the following beneficial effects:

[0051] In the present invention, first, feature extraction is performed on multi-source mine ventilation data. Fast Fourier transform is used to convert time-domain signals into frequency-domain signals and extract spectral features. Then, an evidence body is constructed and the data is fused using the Dempster-Shafer synthesis rule. This can integrate information from multiple sensor nodes and reduce data uncertainty and error. For example, when some sensors produce abnormal data due to environmental interference, the interference can be effectively eliminated by fusing evidence from other sensors, resulting in more reliable confidence distribution data on whether ventilation equipment is normal or faulty.

[0052] Secondly, by using a deep neural network as the foundation and introducing an adaptive reinforcement learning model based on the local perception and weight sharing ideas of convolutional neural networks, combined with a dual-deep Q-network algorithm for strategy updating, the model can better adapt to the complex nonlinear dynamic characteristics of mine ventilation systems. It can learn in real time based on comprehensive state data and generate the optimal preliminary control actions for different mine equipment.

[0053] Finally, by introducing the tunnel structure safety as a dynamic constraint factor and combining it with the improved pigeon flock optimization algorithm, the final control action is obtained. The improved pigeon flock optimization algorithm uses chaotic mapping to initialize the pigeon flock position, which increases the diversity and ergodicity of the initial population and improves the algorithm's search ability. In the position update process, a diversity feedback mechanism is introduced to avoid the algorithm from falling into the local optimal solution and ensure that a better control action combination can be searched. At the same time, by incorporating the tunnel structure safety into the fitness function, the algorithm fully considers the safety of the tunnel structure while optimizing the ventilation effect and energy consumption. When the pigeon position (control action combination) causes the tunnel structure safety to be lower than the preset threshold, the position is adjusted through a reasonable correction amount to ensure the safety of underground workers and equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a method step diagram of a mine intelligent ventilation joint optimization and control method and system proposed by the present invention.

[0055] Figure 2 This is a system block diagram of a mine intelligent ventilation joint optimization and control method and system proposed by the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] Example 1: Figure 1 As shown, the present invention proposes a method for joint optimization and control of intelligent mine ventilation, the method steps including:

[0058] S1: Collect multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy;

[0059] In the mine tunnel, a three-dimensional stereoscopic perception network is constructed with tunnel intersections, curves, and working surfaces as core nodes. Multimodal acoustic wave sensors are deployed in the three-dimensional stereoscopic perception network. The multimodal acoustic wave sensors include

[0060] The horizontal distance between adjacent sensor nodes is , the vertical spacing distance is , forming a regular spatial grid, each sensor node is equipped with a low-power acoustic wave transmitting and receiving module, and each sensor node is set according to the sampling frequency Collect acoustic wave signals in the mine;

[0061] Specifically, the collected acoustic wave signals are discrete signal sequences, which include wind speed data, tunnel structure acoustic signals, and equipment speed signals;

[0062] Wind speed data: Wind speed signals collected by multimodal acoustic sensors can be analyzed to obtain wind speed information at different locations within the mine. For example, at tunnel intersections and bends, wind speeds vary due to changes in tunnel structure. Accurately acquiring wind speed data allows for subsequent decision-making to determine whether ventilation is adequate. Excessively low wind speeds can lead to risks such as gas accumulation.

[0063] Tunnel structure acoustic signals: When sound waves propagate in a tunnel, they are reflected and refracted by tunnel structures (such as intersections and curves). The signals collected by the acoustic sensor contain tunnel structure information.

[0064] Equipment speed signal: Changes in equipment speed will change the characteristics of the sound emitted. The signal collected by the acoustic sensor can indirectly reflect the speed.

[0065] S2: Use trust function theory to integrate and optimize multi-source mine ventilation data to obtain comprehensive status data;

[0066] Trust function theory is a mathematical theory for processing uncertain information. To apply it to mine ventilation decision-making, we first define an identification framework. , identification framework The set of all possible propositions, for mine ventilation systems, identifying the framework Includes information such as "Ventilation equipment normal" and "Ventilation equipment faulty";

[0067] The feature extraction of multi-source mine ventilation data collected from the distributed acoustic wave sensor array is performed, and the time domain signal is transformed into Convert to frequency domain signal ;

[0068] Extract the spectral features of the signal, and assume that the extracted spectral features are ,in, As the feature dimension, the evidence body is constructed based on the extracted spectral features. The evidence body is constructed for different propositions. The evidence body includes: Indicates "ventilation equipment is normal", Indicates "ventilation equipment failure";

[0069] According to the spectrum characteristics , compared with the pre-set fault characteristic threshold, and the basic probability distribution function is determined , for example: if the feature (The amplitude of a certain frequency component of the device operating sound) exceeds the preset normal range threshold , a higher trust is assigned Give the proposition "Failure of ventilation equipment", and ;

[0070] For the spectrum feature data of all sensor nodes, each node constructs an evidence body based on its own collected data and uses the Dempster-Shafer synthesis rule to obtain the trust distribution data;

[0071] The process of obtaining trust distribution data is as follows:

[0072] A fusion algorithm (Dempster-Shafer synthesis rule) is used to fuse the evidence bodies of any two sensor nodes to obtain the basic probability distribution after fusion. , traverse the evidence of all sensor nodes and perform Dempster-Shafer synthesis rule fusion to obtain the corresponding basic probability distribution , For indexing, the basic probability distribution of all sensor nodes is combined together, and finally the trust distribution data about the comprehensive status of the data of each sensor node of mine ventilation is obtained;

[0073] Specifically, in a mine ventilation scenario, multiple sensor nodes acquire data from different locations and angles and construct evidence bodies. Each evidence body represents a local judgment about a certain aspect of the mine ventilation system's status. The purpose of the Dempster-Shafer synthesis rule is to integrate these scattered, local evidence bodies to arrive at a judgment about the mine ventilation system's status that integrates all the evidence information.

[0074] Based on the trust distribution data, according to the order of the propositions in the proposition set, the trust distribution data of each proposition (the trust values ​​after quantification or standardization) are arranged in sequence to form a data sequence, which is expressed as ,in is the data index, data sequence Each data point in represents the trust level of different sensor node states, and comprehensively reflects the state possibility of the data collected by sensor nodes in the mine ventilation system in various aspects. This data sequence is the comprehensive state data.

[0075] Specifically, the trust distribution data obtained after fusion constitutes the data sequence This data sequence arranges the trust values ​​in the order of propositions. Each value integrates the information of all sensor nodes about the corresponding proposition. In the data sequence, the trust value corresponding to the "ventilation equipment failure" proposition of a certain sensor node is obtained after integrating the evidence of each node. It is no longer the judgment of a local node, but an overall assessment of the possibility of ventilation equipment failure after integrating the information of all nodes. This data sequence form intuitively and accurately presents the comprehensive data status of the mine ventilation system.

[0076] S3: Based on the comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal preliminary control actions for different mine equipment;

[0077] The adaptive reinforcement learning model uses a deep neural network (DNN) as its basic architecture and introduces the local perception and weight sharing ideas of a convolutional neural network (CNN) in the hidden layer. It uses the Double Deep Q-Network (DDQN) algorithm to update policies to adapt to the complex nonlinear dynamic characteristics of mine ventilation systems. The adaptive reinforcement learning model consists of an input layer, multiple hidden layers, and an output layer.

[0078] The input layer of the adaptive reinforcement learning model receives comprehensive state data. Let the dimension of the comprehensive state data be , converted into a comprehensive vector representing the dimension, that is, , the number of neurons in the input layer and the dimension of the comprehensive state data Similarly, the comprehensive state data is directly passed to the next hidden layer;

[0079] The hidden layer consists of multiple neurons, which are connected by weights. The activation function is used to perform nonlinear transformation on the output of neurons. The neuron input of the hidden layer is , the output is , the weight matrix is , the bias vector is ,but:

[0080]

[0081] in, is the activation function, which gradually extracts the complex features in the comprehensive state data through nonlinear transformation of multiple hidden layers;

[0082] The specific process of introducing the local perception and weight sharing ideas of convolutional neural networks (CNN) into the hidden layer is as follows:

[0083] In the hidden layer, convolution modules are set at intervals to optimize the convolution operation on the local area of ​​the data, which can be specifically expressed as follows:

[0084] For one-dimensional comprehensive state data, a one-dimensional convolution kernel is used with a size of The convolution kernel slides on the data and performs convolution operation, that is, ,in is the convolution kernel weight, is the bias vector, is the output after convolution optimization;

[0085] Specifically, through this optimization method of local perception and weight sharing, the model's ability to capture local key information in the data is enhanced, while the number of parameters is reduced and training efficiency is improved;

[0086] The output layer outputs the control action scores of different devices. The number of neurons corresponds to the number of device actions under control. Suppose there are Equipment needs to be regulated, each device has possible control actions, the number of neurons in the output layer is , the output layer uses a linear activation function, and the output is a vector representing the probability of the control actions of different devices, that is, ;

[0087] The process of policy update through the dual-depth Q network algorithm is:

[0088] The adaptive reinforcement learning model interacts with the mine ventilation system environment. At each time step, the model receives the current comprehensive state data. As input, the output control action vector , according to the control action vector, perform the corresponding equipment control operation in the mine ventilation system, and then observe the changes in the system state to obtain the new comprehensive state data and the corresponding reward value ;

[0089] Reward Value Defined based on changes in system status. For example, if the energy consumption of the ventilation system decreases and the air quality improves, a positive reward is given; if the ventilation equipment experiences an abnormality or the air quality deteriorates, a negative reward is given;

[0090] The Double Deep Q-Network (DDQN) is used to update the strategy. Based on the Deep Q-Learning (DQL), the Double Deep Q-Network solves the overestimation problem that may occur in DQL and defines the Q function ,in, It is a parameter of the deep neural network, which represents the cumulative reward expected to be obtained by performing actions in the state. The dual-depth Q network includes the current network and the target network;

[0091] The current network selects an action, and the target network evaluates the action value by minimizing the loss function Update current network parameters , the loss function is defined as:

[0092]

[0093] in, , is the reward value, is the discount factor, For the next state, is the current network parameter, are the target network parameters,

[0094] Using the experience replay mechanism, after multiple rounds of training, the experience replay buffer samples are continuously extracted for training and the current network parameters are updated. When the training converges, the current network has learned the best preliminary control strategy under different comprehensive conditions, which is expressed as ,in, The number of indexes for the control strategy;

[0095] For example:

[0096] Suppose a mine ventilation scenario involves three main devices requiring control: the main fan, local fans, and ventilation valves. The main fan has three control actions (increase speed, maintain speed, and reduce speed), the local fans have two control actions (open and close), and the ventilation valves have two control actions (open wide and close narrow). An adaptive reinforcement learning model receives comprehensive status data in real time and outputs the optimal control result. For example, when the input comprehensive status data indicates rising gas concentration and excessive temperature of device 1, the optimal control action output by the current network is to maintain the speed of the main fan, open the local fans, and open the ventilation valves wide. This combination of actions effectively reduces gas concentration and prevents damage to the main fan due to excessive acceleration, achieving optimal control of the mine ventilation system under this condition. In actual operation, as long as the real-time comprehensive status data of the mine ventilation system (including equipment status and air quality indicators) is input into the converged current network, the action combination output by the network will be executed as the optimal control action, thus achieving intelligent control of the mine ventilation system.

[0097] S4: Based on the optimal preliminary control action, the tunnel structure safety factor is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is achieved based on the final control action;

[0098] The process of obtaining the safety degree of tunnel structure is as follows:

[0099] Use safety sensors (such as displacement sensors, stress sensors, etc.) to monitor the deformation and stress parameters of the tunnel, and combine the deformation and stress parameters with the geomechanical model to calculate the tunnel structure safety index. The structural safety index includes the tunnel wall displacement and the stress index of the support structure. According to the tunnel wall displacement and the stress index of the support structure in the structural safety index, the quantified tunnel structure safety value is obtained through the empirical model, that is, the tunnel structure safety index. , and set a tunnel structure safety threshold ;

[0100] Specifically, the collected preliminary safety sensor data is normalized and preprocessed, and the preprocessed displacement and stress data are input into a geomechanical model. The value ranges from 0 to 1, with higher values ​​indicating safer roadway structures;

[0101] The process of introducing the tunnel structure safety as a dynamic constraint factor and combining it with the improved pigeon flock optimization algorithm to obtain the final control action is as follows:

[0102] A pigeon flock optimization algorithm is introduced, which uses chaotic mapping to initialize the position of the pigeon flock to increase the diversity and ergodicity of the initial population. The chaotic mapping position formula is expressed as: ,in, For the control parameters, For the previous position;

[0103] Assume that the size of the pigeon flock in the pigeon flock optimization algorithm is , each pigeon represents a control action combination, the vector dimension is consistent with the number of parameters of the control action, and the position of the initial pigeon group within the parameter value range is randomly generated based on the chaotic map;

[0104] Through a fitness function To evaluate the quality of each pigeon position (i.e., combination of control actions), For the initial regulatory strategy;

[0105] The fitness function comprehensively considers ventilation effect (reduction degree of gas concentration, improvement of air quality), energy consumption and tunnel structure safety factors, and sets the target reduction amount of gas concentration as The actual reduction is , the air quality improvement index is , energy consumption is , then the fitness function can be expressed as:

[0106]

[0107] in, For the safety of the roadway structure, this fitness function combines the roadway structure safety factor and takes the roadway structure safety as a dynamic constraint factor, further increasing the diversity of considerations of the fitness function;

[0108] For each pigeon (i.e., control action combination), calculate its fitness value according to the fitness function, and find the position of the control action combination with the highest fitness, that is, the global optimal position. ;

[0109] Adjust the flight direction of the pigeon flock according to the global optimal position, obtain the flight speed and update the position according to the flight speed, and set The current location of the pigeon is , the speed is , the global optimal position is , the flight speed formula is:

[0110]

[0111] in, is the learning factor;

[0112] The position update formula is: , ensure that the updated position is within the parameter value range;

[0113] A diversity feedback mechanism is introduced during the location update process, specifically:

[0114] Calculating diversity metrics for pigeon flocks , the average distance between pigeons is measured using Euclidean distance. When the diversity index is lower than the preset threshold, it indicates that the pigeon group is too concentrated and is prone to falling into a local optimum. At this time, the positions of the pigeons are randomly disturbed to achieve diversity feedback. The formula is expressed as:

[0115]

[0116] in is the perturbation coefficient, is a standard normally distributed random number, is the new current position;

[0117] Among them, diversity indicators The Euclidean distance between any two pigeons is calculated, and then the Euclidean distance between all pairs of pigeons is added up and divided by the total number of pairs of pigeons.

[0118] Specifically, when the diversity indicator When it is high, it means that the average distance between pigeons in the pigeon flock is large, the difference in the control action combination is large, the pigeon flock is widely distributed in the search space, and can explore different areas more comprehensively, reducing the risk of falling into the local optimal solution. When the diversity index is high, When the value is lower than the preset threshold, it means that the average distance between pigeons is small, the pigeons are too concentrated, and they are all gathered in a local area of ​​the search space. It is easy to fall into the local optimal solution. At this time, it is necessary to increase diversity through random perturbation so that the algorithm can continue to effectively search for the global optimal solution.

[0119] At the same time, the position and speed of the pigeon group are updated considering the safety of the tunnel structure:

[0120] Introduce the correction amount determined according to the tunnel structure safety and control action adjustment rules , when the pigeon position causes the roadway structure safety to be lower than the preset threshold When the correction amount Correct its position. ,in, It is the correction amount determined according to the safety degree of the tunnel structure and the adjustment rules of the control action;

[0121] The process of determining the correction amount based on the roadway structure safety and the control action adjustment rules is as follows:

[0122] Based on the safety threshold of tunnel structure , and the pigeon position Corresponding actual tunnel structure safety Obtain the tunnel structure safety index and define the tunnel structure safety index :

[0123]

[0124] in, The value of reflects the current control action combination (corresponding to the pigeon position ) causes the relative degree of deviation of the tunnel structure safety from the target threshold, The larger it is, the more serious the deviation is;

[0125] Construct the control action adjustment rules, which include the control action fan speed and ventilation valve opening ,Right now , based on the control action (ventilator speed and ventilation valve opening ) Set the fan speed correction value to And the ventilation valve opening correction amount is The fan speed correction value is obtained based on the minimum step length of the fan speed adjustment in the current control action combination, and the ventilation valve opening correction value is obtained based on the minimum step length of the ventilation valve opening adjustment in the current control action combination;

[0126] According to the safety index of tunnel structure and control action adjustment rules to calculate the correction amount ,Right now:

[0127] Get the fan speed correction value :

[0128]

[0129] in, The minimum step size for fan speed adjustment;

[0130] Get ventilation valve opening correction value :

[0131]

[0132] in, The minimum step size for adjusting the ventilation valve opening;

[0133] The fan speed correction is based on And the ventilation valve opening correction amount is Get the final correction amount, expressed as In this way, the adjustment range of the pigeon position (i.e., the control action combination) is related to the safety of the tunnel structure. The more serious the deviation of the safety level, the greater the adjustment range;

[0134] Repeatedly calculate the fitness function, update the position and speed of the pigeon group until the iteration termination condition is met (the maximum number of iterations is reached). ) The final global optimal position The corresponding control action combination introduces the tunnel structure safety as a dynamic constraint factor and combines it with the improved pigeon flock optimization algorithm to obtain the final control action;

[0135] The final control action is converted into specific control instructions and sent to the underground ventilation equipment, controlling the ventilator to operate at the final determined speed and adjusting the ventilation valve to the corresponding opening.

[0136] Example 2: Figure 2 As shown, a mine intelligent ventilation joint optimization and control system includes:

[0137] Data acquisition module: collects multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy;

[0138] Fusion optimization module: Use trust function theory to fuse and optimize multi-source mine ventilation data to obtain comprehensive status data;

[0139] Reinforcement Learning Module: Based on comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal initial control actions for different mine equipment;

[0140] Control optimization module: Based on the optimal preliminary control action, the tunnel structure safety is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is realized based on the final control action.

[0141] In the application, several formulas involved are calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0142] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0143] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for joint optimization and control of intelligent ventilation in mines, characterized in that: The method steps are: S1: Collect multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy; S2: Use trust function theory to integrate and optimize multi-source mine ventilation data to obtain comprehensive status data; S3: Based on the comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal preliminary control actions for different mine equipment; The adaptive reinforcement learning model uses a deep neural network as its basic architecture, introduces the local perception and weight sharing ideas of convolutional neural networks in the hidden layer, and performs strategy updates based on a dual-deep Q-network algorithm to adapt to the complex nonlinear dynamic characteristics of the mine ventilation system. The adaptive reinforcement learning model comprises an input layer, multiple hidden layers and an output layer; S4: Based on the optimal preliminary control action, the tunnel structure safety factor is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is achieved based on the final control action; The process of obtaining the safety degree of the tunnel structure is as follows: Using safety sensors to monitor the deformation and stress parameters of the tunnel, the safety sensors include displacement sensors and stress sensors; The deformation and stress parameters are combined with the geomechanical model to obtain the tunnel structure safety index, and the tunnel structure safety degree is quantified based on the structural safety index through the empirical model. Among them, the improved pigeon flock optimization algorithm is achieved by using chaotic mapping to initialize the pigeon flock position in the pigeon flock optimization algorithm, introducing a diversity feedback mechanism in the pigeon flock position update process, and considering the safety of the tunnel structure; The diversity feedback mechanism calculates the diversity index of the pigeon flock and uses the Euclidean distance to measure the average distance between pigeons. When the diversity index is lower than a preset threshold, the positions of the pigeons are randomly disturbed.

2. A mine intelligent ventilation joint optimization and control method according to claim 1, characterized in that: The process of collecting multi-source mine ventilation data is as follows: Deploy multimodal acoustic wave sensors and space adjacent sensor nodes horizontally at a distance of , the vertical spacing distance is , forming a spatial grid, each sensor node is equipped with a low-power acoustic wave transmitting and receiving module, and each sensor node is set according to the sampling frequency Acoustic wave signals in the mine are collected. The collected acoustic wave signals are discrete signal sequences, which include wind speed data, tunnel structure acoustic signals and equipment speed signals.

3. A mine intelligent ventilation joint optimization and control method according to claim 2, characterized in that: The process of fusion optimization of multi-source mine ventilation data using trust function theory is as follows: Define an identification framework, which includes "ventilation equipment normal" and "ventilation equipment failure"; Feature extraction is performed on multi-source mine ventilation data collected from a distributed acoustic sensor array. The time domain signal is converted into a frequency domain signal through fast Fourier transform, and the spectral characteristics of the signal are extracted. Based on the extracted spectral characteristics, an evidence body is constructed. The evidence body includes both normal ventilation equipment and ventilation equipment failure. Based on the spectrum characteristics, the basic probability distribution function is determined by comparing them with the pre-set fault characteristic threshold. For the spectrum characteristic data of all sensor nodes, each node constructs an evidence body based on its own collected data and uses the Dempster-Shafer synthesis rule to obtain the trust distribution data. Based on the trust distribution data, the trust distribution data of each proposition is arranged in sequence according to the order of the propositions in the proposition set to form a data sequence. This data sequence is the comprehensive status data.

4. A mine intelligent ventilation joint optimization and control method according to claim 3, characterized in that: The process of obtaining trust distribution data using the Dempster-Shafer synthesis rule is as follows: Based on the evidence bodies of any two sensor nodes, the evidence bodies are fused to obtain the fused basic probability distribution. The evidence bodies of all sensor nodes are traversed to perform Dempster-Shafer synthesis rule fusion to obtain the corresponding basic probability distribution. The basic probability distributions of all sensor nodes are combined together to obtain the trust distribution data.

5. The method for joint optimization and control of intelligent ventilation in mines according to claim 1, characterized in that: The process of obtaining the preliminary control action is as follows: The input layer of the adaptive reinforcement learning model receives the comprehensive state data, converts it into a comprehensive vector representing the dimension, and then transmits it to the hidden layer; The hidden layer consists of multiple neurons, which are connected by weights. The activation function is used to perform nonlinear transformation on the output of the neurons. The local perception and weight sharing ideas of convolutional neural networks are introduced. Convolution modules are set at intervals to optimize the convolution operation on local areas of the data. The output layer outputs the control action scores of different devices. The number of neurons corresponds to the number of device actions controlled, and the output layer outputs the probability of control actions of different devices. Strategy update is performed through the dual-depth Q network algorithm. The dual-depth Q network includes the current network and the target network. The Q function is defined. The Q function includes the parameters of the deep neural network. The current network is used to select actions, and the target network evaluates the value of actions. The current network parameters are updated by minimizing the loss function and adopting the experience revisit mechanism. After multiple rounds of training, the current network parameters are updated. When the training converges, the optimal preliminary control strategy under different comprehensive states is obtained.

6. A mine intelligent ventilation joint optimization and control method according to claim 1, characterized in that: The process of obtaining the final control action by introducing the tunnel structure safety as a dynamic constraint factor and combining it with the improved pigeon flock optimization algorithm is as follows: Introducing the pigeon flock optimization algorithm and using chaos mapping to initialize the pigeon flock position Assume the size of the pigeon flock is , each pigeon represents a combination of control actions, the vector dimension is consistent with the number of parameters of the control action, and the position of the initial pigeon group within the parameter value range is randomly generated; A fitness function is used to evaluate the pros and cons of each control action combination. For each control action combination, its fitness value is calculated according to the fitness function, and the position of the control action combination with the highest fitness is found, which is the global optimal position. Adjust the flight direction of the pigeon flock according to the global optimal position, obtain the flight speed, and update the position according to the flight speed; A diversity feedback mechanism is introduced during the position update process. The diversity index in the diversity feedback mechanism is obtained by calculating the Euclidean distance between any two pigeons, then adding up the Euclidean distances between all pairs of pigeons and dividing the sum by the total number of pairwise combinations of pigeons. According to the correction amount determined by the tunnel structure safety and the control action adjustment rules, when the pigeon's position causes the tunnel structure safety to be lower than the preset safety threshold, its position is corrected by the correction amount; Repeatedly calculate the fitness function, update the position and speed of the pigeon group until the iteration termination condition is met and the maximum number of iterations is reached. The global optimal position is finally obtained. The corresponding control action combination is the final control action.

7. The method for joint optimization and control of intelligent ventilation in mines according to claim 1, characterized in that: The process of determining the correction amount according to the tunnel structure safety and the control action adjustment rules is as follows: The tunnel structure safety index is obtained based on the tunnel structure safety threshold and the actual tunnel structure safety corresponding to the pigeon position; Construct a control action adjustment rule, which includes the fan speed and ventilation valve opening. Set the fan speed correction value to be and the ventilation valve opening correction value to be. The fan speed correction value is obtained based on the minimum step length of the fan speed adjustment considered in the current control action combination. The ventilation valve opening correction value is obtained based on the minimum step length of the ventilation valve opening adjustment considered in the current control action combination: The final correction amount is determined based on the fan speed correction amount and the ventilation valve opening correction amount.

8. A mine intelligent ventilation joint optimization and control system, using the method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: collects multi-source mine ventilation data using distributed multimodal acoustic sensors with a grid deployment strategy; Fusion optimization module: Use trust function theory to fuse and optimize multi-source mine ventilation data to obtain comprehensive status data; Reinforcement Learning Module: Based on comprehensive state data, an adaptive reinforcement learning model is used to obtain the optimal initial control actions for different mine equipment; Control optimization module: Based on the optimal preliminary control action, the tunnel structure safety is introduced as a dynamic constraint factor and combined with the improved pigeon flock optimization algorithm to obtain the final control action, and underground ventilation control is realized based on the final control action.

Images