A mine ventilation optimization method and system based on an edge computing architecture

Abstract

The application discloses a kind of mine ventilation optimization method and system based on edge computing architecture, and optimization method includes the following steps: gas concentration information, air velocity information and temperature information perceived to the environmental perception unit are corrected and normalized processing;Characteristic vector is extracted to normalized data, the weighted Manhattan distance of characteristic vector and window center is calculated, and according to the weighted distance matching of characteristic vector and preset working condition characteristic matrix, in combination with historical window frequency statistics and time consistency determination, output working condition label;Build digital twin ventilation network model, to design energy consumption minimization, air volume distribution uniformity maximization, equipment life equalization as optimization goal, according to the output working condition label, build total target optimization function;Solve total target optimization function, obtain optimal control instruction, and issue to underground fan and air door, and dynamically adjust control instruction according to gas concentration information, air velocity information and temperature information obtained by environmental perception unit in real time.

Description

Technical Field

[0001] This invention relates to the intersection of industrial Internet of Things, edge computing, and intelligent control technology for mine safety, and particularly to a method and system for optimizing mine ventilation based on an edge computing architecture. Background Technology

[0002] Mine ventilation systems are the "circulatory system" of mine safety production. Their core task is to establish a stable and reasonable airflow network underground, continuously deliver fresh air from the surface to all work sites, and at the same time dilute and exhaust toxic and harmful gases, dust, heat and moisture underground, creating a safe and healthy working environment for personnel and preventing serious accidents such as gas outbursts and fires.

[0003] Most mines still use traditional ventilation control methods: 1. Centralized control mode based on fixed thresholds: This system has a central control room on the ground and relies on a sensor network distributed throughout the well to collect data. All data is transmitted to the ground server via industrial Ethernet or bus, and the control system presets fixed action thresholds. This method is essentially a "reactive response" and has serious lag. In addition, in order to ensure that ventilation requirements are still met under the worst conditions, the system usually adopts a "full air volume, always open" operation strategy, which causes the main fan and auxiliary fan to be in a high-energy-consuming state for a long time, resulting in huge power waste.

[0004] 2. Schedule-based program control mode: Different ventilation schemes are preset according to the shifts and work patterns of mine production, and switched according to the schedule. This method is difficult to adapt to dynamic changes in the underground environment, has extremely poor flexibility, and its safety depends on the conservatism of the preset scheme, resulting in serious energy waste.

[0005] 3. Intelligent ventilation control mode based on "cloud-edge" structure: Its architecture mostly adopts a centralized processing paradigm of "sensor-transmission network-cloud / central server-actuator". This method has strong network dependence and poor robustness, making it difficult to cope with the complex environment downhole; the central server is under great pressure, requiring high computing power, storage bandwidth and data processing software; the response latency is high, and the inherent physical distance and network protocol delays make it unable to meet the needs of rapid response.

[0006] In summary, existing ventilation control systems struggle to simultaneously achieve a balance between safety, real-time performance, reliability, and cost-effectiveness. The rise of edge computing technology offers an ideal technological solution to this problem. It moves computing power from the network core to the network edge, closer to the data source and actuators, perfectly meeting the urgent needs of mine ventilation control for low latency, high reliability, and local intelligent decision-making.

[0007] This invention patent adopts this emerging edge computing paradigm, and on this basis, it has carried out detailed algorithm design for the edge layer and the cloud, which can realize dynamic intelligent ventilation effect based on sensor data processing, anomaly pre-detection, working condition identification, and strategy optimization. Summary of the Invention

[0008] This solution addresses the problems and needs raised above by proposing a mine ventilation optimization method and system based on an edge computing architecture. Due to the adoption of the following technical features, it can achieve the above technical objectives and bring about several other technical effects.

[0009] One objective of this invention is to propose a mine ventilation optimization method based on an edge computing architecture, comprising the following steps: S10: Deploy edge computing nodes and environmental sensing units at key locations underground, and correct and normalize the gas concentration, wind speed and temperature information sensed by the environmental sensing units. S20: Extract feature vectors from normalized data, calculate the weighted Manhattan distance between the feature vectors and the center of the window, match the feature vectors with the weighted distance of the preset working condition feature matrix, and combine historical window frequency statistics and time consistency judgment to output working condition labels. ; S30: Construct a digital twin ventilation network model with the optimization objectives of minimizing design energy consumption, maximizing airflow distribution uniformity, and balancing equipment lifespan, based on the output operating condition labels. Construct the overall objective function; S40: The weighted particle swarm optimization algorithm is used to solve the overall objective function and obtain the optimal control command. The commands are then sent to the underground ventilation fans and dampers, and the control instructions are dynamically adjusted based on the real-time gas concentration, wind speed, and temperature information obtained by the environmental sensing unit. .

[0010] In addition, the mine ventilation optimization method based on edge computing architecture according to the present invention may also have the following technical features: In one example of the present invention, step S10 involves correcting and normalizing the gas concentration, wind speed, and temperature information sensed by the environmental sensing unit, including the following steps: S11: Uses a fixed frequency Collect gas concentration, wind speed, and temperature data, and based on the previous... The mean of the data and range Dynamically calculate the current time. The anomaly detection threshold range, where the anomaly detection threshold is... formula: In the formula, This represents the size of the sliding window. For the front Moving average of the data; For the front The range of the data; This is the threshold coefficient; This is the range conversion factor; S12: When new sampling data Check the sampled data during input. Does it fall within the threshold? If within the upper and lower limits, Within this range, the data is considered normal; if the sampled data... Data outside this range is considered outlier and will be corrected using a correction formula, which is: S13: The corrected sampling data Normalization processing mapping to Interval, to eliminate the influence of dimensions and obtain Time-normalized output The formula for normalization is: In the formula, For a set of buffers; The minimum value in the buffer; This represents the maximum value in the buffer.

[0011] In one example of the present invention, step S20 includes the following steps: S21: Feature extraction is performed on the normalized data by solving the feature extraction formula, which is: In the formula, For feature vectors; For sliding statistics windows; This represents the mean value within the window. The standard deviation within the window; S22: Predict the next H time steps based on current features; where the time series prediction formula is... for: In the formula, This is the trend decay constant; To predict the step size; for Time-normalized output data; S23: Detecting abnormal data based on feature spatial distance; S24: Output five-dimensional feature vector As input to the working condition identification module, the output is a working condition category label. It identifies five working conditions—unknown, normal, maintenance, gas outburst, and fire emergency—by using distance calculation and rule-based judgment.

[0012] In one example of the present invention, step S23 includes the following steps: S231: Calculate the weighted Manhattan distance of data features Its expression is: In the formula, These are the weight vectors, corresponding to the five dimensions of the feature vectors; For feature vectors; To indicate the number within the history window The average of the features; S232: Adaptive threshold setting Its expression is: In the formula, These are the mean and standard deviation of the historical feature distances, respectively; This is the threshold coefficient; Among them, the threshold coefficient The expression is: In the formula, The basic threshold coefficient; To adjust the amplitude coefficient; This is the sensitivity coefficient; This serves as a reference for the maximum level of fluctuation. To measure the fluctuation value of the anomaly detection results themselves; in, The expression is: S233: Determine whether data features are abnormal based on confidence level, where confidence level... The expression is: Among them, the confidence level evaluation directly reflects the proportion of abnormal distance to the total distance, reflecting the relative degree of abnormality. hour, It is at the threshold boundary; when hour, , indicating a high degree of abnormality; when hour, This indicates a highly normal level.

[0013] In one example of the present invention, step S24 includes the following steps: S241: Perform data feature annotation processing. The expression is: In the formula, For the first Historical statistical mean and standard deviation of each feature; To prevent constants with zero denominators, here ; S242: Based on the anomaly type feature comparison table, determine the gas outburst and fire conditions, including emergency determination for gas outbursts. The expression is: Fire emergency determination Expressed as: S243: Determine unknown working conditions, normal generation, and maintenance status based on the working condition determination formula, whereby the working condition label... The expression for the determination formula is: In the formula, The criteria are based on weighted distance matching degree, frequency statistics, and time consistency. The score threshold; S244: When emergency features such as gas outburst or fire are detected, the corresponding emergency condition is directly determined; if no emergency condition is detected, then... The function scores the conditions and assigns a condition label to the corresponding working conditions; if no emergency condition is found and the output of the scoring function is below the scoring threshold, it indicates that an unknown working condition has been detected.

[0014] In one example of the present invention, step S30 includes the following steps: S31: Define decision variables : In the formula, For the first The set values ​​for each ventilation control device; among them, for the fan, , For operating frequency; for dampers, , Percentage opening; For the total number of devices, In the formula, Where N is the total number of fans and N is the total number of dampers; S32: Design the overall objective optimization function F with the three optimization objectives of minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan; the expression of the overall objective optimization function F is: .

[0015] In one example of the present invention, in step S32, a general objective optimization function F is designed with three optimization objectives: minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan; specifically, it includes the following steps: S321: Construct the energy consumption minimization objective function: where, Wind turbine energy consumption model The expression is: In the formula, For wind turbine Base power consumption under no-load conditions; The coefficient for the first-order term represents the linear losses of the wind turbine; The coefficient for the quadratic term represents the mechanical losses of the fan; The coefficient for the cubic term represents the aerodynamic power of the fan; For wind turbine Operating frequency; This model replaces the traditional one with a complete trinomial. The model includes mechanical losses and motor iron losses; Damper energy consumption model The expression is: In the formula, For air damper The drag coefficient, ; For air damper The degree of opening, ; Energy minimization objective function The expression is: S322: Construct an objective function to maximize the uniformity of airflow distribution based on digital twins, where, Digital twin-based air volume prediction model The expression is: In the formula, Regarding the damper opening The impedance matrix; Regarding the power of the wind turbine Wind turbine power vector; Objective function for maximizing airflow distribution uniformity based on digital twin Its expression is: In the formula, The number of key roadways and working faces; For tunnels based on digital twin CFD models Predict air volume; This represents the average air volume for the entire mine. ; S323: Constructing the objective function for equipment life balancing Its expression is: In the formula, For wind turbine Rated power; respectively wind turbine Upper and lower limits of frequency; For air damper The rated opening degree; They are air dampers The upper and lower limits of the opening; S324: Edge-side condition labeling The dynamic weight adjustment overall objective function F is expressed as: In the formula, These are the edge-side working condition labels. Dynamic weights.

[0016] In one example of the present invention, step S40 specifically includes the following steps: S41: Perform initialization settings, generate particle positions within feasible locations. That is, the particles are combinations of device set values, and a fitness evaluation is performed to match the overall objective optimization function. The fitness function is... The expression is: In the formula, These correspond to the energy consumption component, the uniformity component, and the lifetime component, respectively, which are the three optimization objectives. Individual optimal position update rule The expression is: In the formula, The particle number represents a different candidate solution, and each particle represents a possible ventilation strategy. The position vector parameter represents the particle's position, i.e., the particle's position vector. In the The position within a cycle corresponds to the value vector of the ventilation equipment. Represents the individual's optimal vector parameters, storing particles. The optimal solution found during the search process; initial stage Iterative updates are performed; if the updated position is better, then it is updated to... Otherwise, keep it; S42: Initialization Each particle is randomly generated within a feasible region. This describes a device's configuration scheme: In the formula, Let be the vector space of the particles; This is the device lower bound vector; The upper limit vector of devices; The fitness function of each particle is obtained based on the optimization function in the multi-objective optimization decision module. The expression is: Since there is more than one piece of equipment underground when applying to real-world mining scenarios, a core memory variable needs to be added. Used to characterize the global historical best position Its expression is: In the formula, This is a weighted fitness function; The number of particles; This represents the current iteration number; S43: At the start of the algorithm, all particles are randomly initialized with their positions, their fitness is calculated, and the particle with the lowest fitness is found. And set the initial global optimal position. Its expression is: After each iteration from initialization, the particle position is updated, and the global optimal position variable is... It will compare the new positions generated by all particles, and if a better position is found, it will be updated. Otherwise, it remains unchanged; among them, The update mechanism expression is: The algorithm will search for all particles within the region, that is: S44: Based on the individual optimal position update rule and the global optimal position update rule, formulate the particle update rule within the region to characterize the change in device state; among which, the particle update rule The expression is: In the formula, This is the inertial term, meaning it maintains the search direction from the previous moment; This represents the individual experience term, meaning that the particle learns from the individual's historical best position, and the particle optimizes instructions based on the device's historical operating experience. This represents the social learning term, where each particle learns from the historical best position found by the entire population. Particles draw on the optimization experience of other devices or regions, even if a particle's own historical best position is not the best. Even if it's not ideal, you can still learn from the best experiences of the group. Obtain from; After the algorithm completes its iteration rounds, The final value is the global optimal solution: In the formula, This represents the maximum number of iterations. Final algorithm output The optimized control commands are then sent to the terminal fans and dampers.

[0017] Another objective of this invention is to propose a mine ventilation optimization system based on an edge computing architecture, comprising: a perception and execution layer, an edge computing layer, and a cloud platform layer, wherein... The sensing and execution layer includes an environmental sensing unit and an execution control unit. The environmental sensing unit is configured to sense gas concentration, wind speed, and temperature information within the tunnel. The execution control unit is configured to precisely control the airflow within the tunnel by adjusting the frequency of the fan and damper. The edge computing layer includes multiple edge computing nodes, each edge computing node including a data acquisition and preprocessing module and an intelligent analysis module. The data acquisition and preprocessing module is configured to acquire gas concentration information, wind speed information, and temperature information and perform normalization processing to form a normalized data set. The intelligent analysis module is configured to extract features from the normalized data set and perform time-series prediction, anomaly detection, and identification of the current operating condition based on the extracted features. The cloud platform layer is configured to receive edge-side data from abnormal edge computing nodes, perform simulations based on digital twins, construct a global ventilation network model, and establish a multi-objective optimization model with the goals of achieving safety standards, minimum energy consumption, most uniform airflow distribution, and minimum equipment loss; solving this optimization model yields the optimal control command. The specific set values ​​of each fan and damper are obtained, and the optimal control command is sent to the execution control unit to achieve ventilation optimization.

[0018] In one example of the present invention, the cloud platform layer includes: The multi-objective optimization decision module is configured to design a total objective optimization function with the objectives of minimizing energy consumption, maximizing the uniformity of air volume distribution, and balancing equipment lifespan. The weighted particle swarm optimization module is configured to use the weighted particle swarm optimization algorithm to optimize the objective function and obtain the best control scheme. The control command issuing module is configured to issue the acquired optimal control scheme to the execution control unit.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention adds an edge computing layer to the traditional "cloud-edge" structure, enabling some core computing power to be deployed at the edge, realizing hierarchical deployment and dynamic scheduling of computing power. The edge layer and the cloud achieve dynamic collaboration. Under normal circumstances, computing power only stays at the edge, and cloud computing power is called in case of an anomaly. This significantly saves transmission bandwidth and cloud platform computing resources, reduces overall energy consumption and operation and maintenance costs, and at the same time, the edge to end uses a local area network instead of a wide area network in the traditional mode, which improves the robustness of the network structure.

[0020] 2. The present invention designs a lightweight multi-dimensional feature fusion and adaptive anomaly recognition algorithm on the edge side, including real-time data preprocessing based on dynamic window, multi-task feature extraction, weighted Manhattan distance anomaly detection and working condition recognition model. It comprehensively represents the ventilation system status through five-dimensional feature vectors, and combines sliding window statistics and time consistency verification to achieve accurate identification of anomaly types and continuous and stable identification of working conditions. It has a rapid response capability, especially for emergency working conditions such as gas outburst and fire.

[0021] 3. This invention designs a multi-objective optimization and operating condition adaptive strategy in the cloud, and proposes a multi-objective ventilation optimization model based on weighted and particle swarm optimization algorithms. It can dynamically adjust the optimization weights (energy consumption, uniformity, equipment life) according to the operating condition type reported from the edge, realize intelligent switching from "energy saving mode" to "safety priority mode", and improve the overall safety and economy of the system.

[0022] 4. Under the premise of ensuring mine ventilation safety, this invention establishes a three-objective optimization function, and effectively reduces the energy consumption of the fan, extends the equipment life, and balances the underground air volume distribution through optimized control strategies, thus achieving the dual goals of safe production and energy conservation.

[0023] The preferred embodiments of the invention will be described in more detail below with reference to the accompanying drawings, so as to facilitate an understanding of the features and advantages of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. The drawings are merely illustrative of some embodiments of the present invention and are not intended to limit the scope of the present invention to all embodiments.

[0025] Figure 1 This is a flowchart of the intelligent optimization algorithm for mine ventilation according to an embodiment of the present invention; Figure 2 This is an architecture diagram of a mine ventilation optimization system based on an edge computing architecture according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The same reference numerals in the drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, “an” or “a” and similar terms do not necessarily indicate a quantity limitation. Terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships; these relative positional relationships may change accordingly when the absolute position of the described object changes.

[0028] According to a first aspect of the present invention, a mine ventilation optimization method based on an edge computing architecture, such as... Figure 1 As shown, it includes the following steps: S10: Deploy edge computing nodes and environmental sensing units at key underground locations, and correct and normalize the gas concentration, wind speed, and temperature information sensed by the environmental sensing units. Specifically, the environmental sensing units include intrinsically safe or explosion-proof gas concentration sensors, wind speed sensors, and temperature sensors. These sensors are deployed with a certain spatial density and topology to form a sensing network covering the entire mine. Edge computing nodes are typically located in locations with good network and power supply conditions, such as mining area substations and roadway intersections. Each edge computing node is responsible for managing all sensing and execution devices within a logical area (such as a coal face, a tunneling head, or a zoned roadway).

[0029] S20: Extract feature vectors from normalized data, calculate the weighted Manhattan distance between the feature vectors and the center of the window, match the feature vectors with the weighted distance of the preset working condition feature matrix, and combine historical window frequency statistics and time consistency judgment to output working condition labels. ; S30: Construct a digital twin ventilation network model with the optimization objectives of minimizing design energy consumption, maximizing airflow distribution uniformity, and balancing equipment lifespan, based on the output operating condition labels. Construct the overall objective function; S40: The weighted particle swarm optimization algorithm is used to solve the overall objective function and obtain the optimal control command. The commands are then sent to the underground ventilation fans and dampers, and the control instructions are dynamically adjusted based on the real-time gas concentration, wind speed, and temperature information obtained by the environmental sensing unit. .

[0030] This optimization method adds an edge computing layer to the traditional "cloud-edge" structure, enabling some core computing power to be deployed and dynamically scheduled at the edge. The edge layer and the cloud achieve dynamic collaboration. Under normal circumstances, the computing power stays only at the edge, and in case of an anomaly, the cloud computing power is called upon, which greatly saves transmission bandwidth and cloud platform computing resources, reduces overall energy consumption and operation and maintenance costs. At the same time, the edge-to-end uses a local area network instead of a wide area network in the traditional mode, which improves the robustness of the network structure.

[0031] The optimization method designs a lightweight multi-dimensional feature fusion and adaptive anomaly recognition algorithm on the edge side, including real-time data preprocessing based on dynamic windows, multi-task feature extraction, weighted Manhattan distance anomaly detection and working condition recognition model. It comprehensively represents the ventilation system status through five-dimensional feature vectors, and combines sliding window statistics and time consistency verification to achieve accurate identification of anomaly types and continuous and stable identification of working conditions. It has a rapid response capability, especially for emergency working conditions such as gas outbursts and fires.

[0032] This optimization method employs a cloud-based multi-objective optimization and operating condition adaptive strategy. It proposes a multi-objective ventilation optimization model based on weighted and particle swarm optimization algorithms, which can dynamically adjust the optimization weights (energy consumption, uniformity, and equipment lifespan) according to the operating condition types reported from the edge. This enables intelligent switching from "energy-saving mode" to "safety-first mode," thereby improving the overall safety and economy of the system.

[0033] This optimization method establishes a three-objective optimization function while ensuring mine ventilation safety. By optimizing the control strategy, it effectively reduces the energy consumption of the fan, extends the equipment life, and balances the distribution of underground air volume, thus achieving the dual goals of safe production and energy conservation.

[0034] In one example of the present invention, step S10 involves correcting and normalizing the gas concentration, wind speed, and temperature information sensed by the environmental sensing unit, including the following steps: S11: Uses a fixed frequency Collect gas concentration, wind speed, and temperature data, and based on the previous... The mean of the data and range Dynamically calculate the current time. The anomaly detection threshold range, where the anomaly detection threshold is... formula: In the formula, This represents the size of the sliding window. For the front Moving average of the data; For the front The range of the data; This is the threshold coefficient; This is the range conversion factor; S12: When new sampling data Check the sampled data during input. Does it fall within the threshold? If within the upper and lower limits, Within this range, the data is considered normal; if the sampled data... Data outside this range is considered outlier and will be corrected using a correction formula, which is: S13: The corrected sampling data Normalization processing mapping to Interval, to eliminate the influence of dimensions and obtain Time-normalized output The formula for normalization is: In the formula, For a set of buffers; The minimum value in the buffer; This represents the maximum value in the buffer.

[0035] In other words, a lightweight adaptive sensor data acquisition and preprocessing algorithm is used to process the raw data collected by various sensors. This algorithm has a relatively simple logical framework and low computational load, making it suitable for deployment on edge servers. Here, we use gas data as an example to illustrate the specific algorithm framework and processing logic.

[0036] Let a sensor's raw data stream be Create a buffer BUF of size 20, and process each new data return. Store the data in a buffer BUF as the initial data for the algorithm, and then retrieve the 10 most recent values ​​from BUF as the window. This step means setting a fixed length for each sensor. A fixed window that stores the most recent If the current window is full, remove the oldest data. Then, calculate the average value. And the range (the difference between the maximum and minimum values) For the repair and reconstruction of abnormal data, this invention employs the standard deviation threshold method. The rule is used to replace the dual-index anomaly detection method of Z-score + data change rate. This is because during normal production, the downhole ventilation conditions and temperature fluctuations are relatively stable. Whether it is a gas concentration sensor or a wind speed, pressure, or temperature sensor, the data from a single sensor basically follows a normal distribution, that is, about 90% or more of the data fall at the mean. Within the normal range, but under extreme and sudden conditions (such as gas outbursts or sudden temperature increases), the range of sensor data changes far exceeds [the normal range]. The interval is easily detectable, which aligns with the unique characteristics of the mining environment; simultaneously, this method effectively reduces computational load, meeting the lightweight requirements of edge computing. It should be noted that for a very small number of non-normally distributed data points, false positives or false negatives may occur. To compensate for this deficiency, an absolute threshold can be set to enforce constraints, such as for methane concentration. ≥ Upper limit threshold When this happens, the data is directly identified as abnormal. This method works in parallel with statistical anomaly detection.

[0037] The core of this algorithm consists of a quintuple, namely ( These five elements constitute the core mathematical logic of edge data acquisition and preprocessing. The following is an introduction to the formulas and parameters involved in the algorithm: Fixed frequency sampling formula: In the formula, For the first The timestamp of the sample; This is the starting sampling time; The sampling interval; This formula defines how the timestamps for data collection are generated, ensuring that the data is collected at a fixed frequency. (e.g., 5Hz) is sampled to provide a unified time reference for subsequent sliding window and timing analysis.

[0038] Anomaly detection threshold formula: Based on the previous The mean of the data and range Dynamically calculate the current time. The anomaly detection threshold range. It provides the basis for the next formula (the revised formula), when new sampled data... The system will check during input. Does it fall within the threshold? If within the upper and lower limits, Within this range, the data is considered normal; if Data outside this range is considered abnormal and will be corrected using a correction formula.

[0039] Correction formula: This formula is based on the anomaly detection threshold formula for judgment. When If the data exceeds the threshold range, historical averages should be used for replacement to ensure data quality and avoid outliers interfering with subsequent analysis. Failure to use correction formulas for noise filtering may result in false alarms, missed detection of real risks, and increased computational burden. Instantaneous outliers (such as sudden jumps in gas concentration) may be misjudged as precursors to gas outbursts; noise may mask true trends (such as a slowly rising gas concentration being obscured by random fluctuations), preventing the anomaly detection model from identifying early signs in a timely manner and delaying optimal response; abnormal data may cause frequent triggering of cloud collaboration at the edge layer, violating the design principle that computing power should only remain at the edge under normal operating conditions, triggering unnecessary cloud computing power calls and control command adjustments, wasting resources and interfering with normal production.

[0040] Normalization process: Based on buffer The maximum and minimum values ​​in the data will be corrected. Mapped to The interval is used to eliminate the influence of dimensions, so that data from different sensors can be uniformly input into the AI ​​model, thereby improving the model's stability and generalization ability.

[0041] This algorithm is adaptable to various types of sensors. Its adaptability and dynamism are demonstrated through moving window statistics, and its recursive calculation logic ensures constant processing latency, guaranteeing its real-time performance. Through lightweight optimization, the algorithm significantly improves computational efficiency, reducing latency from hundreds of milliseconds to tens or even hundreds of milliseconds.

[0042] In one example of the present invention, step S20 includes the following steps: S21: Feature extraction is performed on the normalized data by solving the feature extraction formula, which is: In the formula, For feature vectors; For sliding statistics windows; This represents the mean value within the window. The standard deviation within the window; S22: Predict the next H time steps based on current features; where the time series prediction formula is... for: In the formula, This is the trend decay constant; To predict the step size; for Time-normalized output data; S23: Detecting abnormal data based on feature spatial distance; S24: Output five-dimensional feature vector As input to the working condition identification module, the output is a working condition category label. It identifies five working conditions—unknown, normal, maintenance, gas outburst, and fire emergency—by using distance calculation and rule-based judgment.

[0043] By solving the feature extraction formula, we can obtain information about... eigenvectors As can be seen, this feature vector is a five-dimensional vector, with each dimension representing: short-term mean feature, instantaneous rate of change, short-term volatility, periodic encoding, and rate of change compression.

[0044] It should be noted that the time-series prediction branch, the anomaly detection branch, and the operating condition identification branch are all based on the same feature extraction algorithm.

[0045] Anomaly detection based on feature space distance: The anomaly detection branch is based on multi-dimensional distance and an adaptive threshold update mechanism. The core logic is that under normal operating conditions, sensor data features should cluster in a specific region in the feature space, while abnormal data will significantly deviate from this region. This anomaly detection is also based on a sliding window about the feature vector. To match the previous module and the data acquisition and preprocessing module, the window size is [value missing]. Each time a new feature vector obtained from feature extraction is added to the window, it is added. If the window is full, the oldest feature vector is removed. And solve for the center of the mean of the vectors in the window. Correspondingly, a fixed length is set. The historical distance queue is updated accordingly after the window is updated, further updating the mean and standard deviation of the distances in the current queue. This invention uses weighted Manhattan distance instead of Euclidean distance, which can effectively reduce the amount of computation.

[0046] Anomaly Type Identification: Clearly, the anomaly detection section above is based on the feature vector set within a sliding window. The essence of anomaly type identification is to use feature vectors to represent different types of abnormal states and to infer possible anomaly types based on changes in different feature dimensions within the feature vectors. Therefore, it is necessary to reiterate the meaning of each dimension in the five-dimensional feature vector: short-term mean feature, instantaneous rate of change, short-term volatility, periodic encoding, and rate of change compression.

[0047] This invention designs an anomaly feature extraction library targeting the four most common anomaly types in underground coal mines, specifically including three types of sensor data anomalies and sensor fault identification, as shown in Table 1: Table 1 is a comparison table of abnormal type characteristics. In one example of the present invention, step S23 includes the following steps: S231: Calculate the weighted Manhattan distance of data features, its expression is: In the formula, These are the weight vectors, corresponding to the five dimensions of the feature vectors; For feature vectors; To indicate the number within the history window The average of the features; S232: Adaptive threshold setting, its expression is: In the formula, These are the mean and standard deviation of the historical feature distances, respectively; This is the threshold coefficient; Among them, the threshold coefficient The expression is: In the formula, The basic threshold coefficient; To adjust the amplitude coefficient; This is the sensitivity coefficient; This serves as a reference for the maximum level of fluctuation. To measure the fluctuation value of the anomaly detection results themselves; in, The expression is: S233: Determine whether data features are abnormal based on confidence level, where the expression for confidence level is: Among them, the confidence level evaluation directly reflects the proportion of abnormal distance to the total distance, reflecting the relative degree of abnormality. hour, It is at the threshold boundary; when hour, , indicating a high degree of abnormality; when hour, This indicates a highly normal level.

[0048] It should be noted that the anomaly detection logic based on weighted Manhattan distance and adaptive threshold is as follows: Judgment criteria based on confidence evaluation Essentially, it determines whether the current data is within a normal range by comparing a weighted Manhattan distance with an adaptive threshold. If , it means the current data is within the normal range, that is, there is no anomaly at the current moment; if If this occurs, it indicates that the current state has significantly deviated from the normal pattern, and the subsequent anomaly type identification and cloud-based collaborative decision-making module needs to be triggered.

[0049] In one example of the present invention, step S24 includes the following steps: S241: The data features are labeled, and the expression is: In the formula, For the first Historical statistical mean and standard deviation of each feature; To prevent constants with zero denominators, here ; S242: Based on the anomaly type feature comparison table, determine the gas outburst and fire conditions. The emergency determination expression for a gas outburst is as follows: The emergency fire assessment is expressed as follows: S243: Determine unknown operating conditions, normal generation, and maintenance states based on the operating condition determination formula, wherein the expression of the operating condition determination formula is: In the formula, The criteria are based on weighted distance matching degree, frequency statistics, and time consistency. The score threshold; S244: When emergency features such as gas outburst or fire are detected, the corresponding emergency condition is directly determined; if no emergency condition is detected, then... The function scores the conditions and assigns a condition label to the corresponding working conditions; if no emergency condition is found and the output of the scoring function is below the scoring threshold, it indicates that an unknown working condition has been detected.

[0050] For example, the following is a matrix example of historical data for four operating conditions: normal generation, maintenance, gas outburst, and fire: Furthermore, it is necessary to measure the difference between the current system state and the known standard operating condition in the multi-dimensional feature space, that is, the meaning of the five dimensions in the corresponding feature vector. Dimension 1 means the average level of the data, reflecting the overall working state; Dimension 2 means the dynamic rate of change of the data, reflecting the changing trend of the system; Dimension 3 means the volatility of the data, reflecting the stability of the data and the stability of the system; Dimension 4 means the periodic regularity of the data, reflecting the working rhythm; Dimension 5 means the trend development direction, reflecting the future development trend. Based on the importance of the five dimensions, take... Same as above.

[0051] Weighted distance calculation: In the formula, The first quantized data of the sensor One eigenvalue; Based on The first working condition One eigenvalue; For the first The weight coefficients of each feature, ; For the current state and the first For different operating conditions, the smaller the value, the closer they are. Matching degree calculation based on weighted distance: In the formula, To match the width parameter (standard deviation) and control the rate of similarity decrease, here... ; The higher the matching degree, the closer it is to the actual working conditions.

[0052] At continuous points in time, the system's operating conditions should exhibit a certain degree of continuity and stability, without frequent jumps or discrepancies. For example, under normal production conditions, the data from each sensor should remain stable for several hours or even days; under maintenance conditions, the data from each sensor should switch in a planned and regular manner, with the duration potentially remaining stable within a few hours; under gas outburst conditions, the concentration sensor data should surge rapidly and then slowly decrease, with the duration within a few minutes or hours; under fire conditions, the temperature sensor data should gradually increase and remain continuous, with the duration ranging from a few minutes to a few hours.

[0053] Therefore, this invention designs a sliding window statistical algorithm for time consistency verification. Window length As mentioned earlier, this is to match the regular sampling frequency (5Hz) and to align the historical operating condition label sequence with the timestamp of the sensor data, avoiding the inclusion of early obsolete data or reducing the number of effective data samples, which would lead to a decrease in noise resistance.

[0054] Historical window statistics: Operating conditions within the historical window Frequency statistics: In the formula, The historical operating condition label indicates The type of operating condition at any given time; Time consistency determination: In the formula, Here, the consistency threshold represents the minimum frequency required to determine consistency. That is, at least 70% of historical points support the current operating condition, when At that time, at least 7 historical points are required to support this. In summary, the decision logic formula based on weighted distance matching degree, frequency statistics, and time consistency judgment can be derived as follows: ) In the formula, Assign weights respectively, for example, ; As can be seen, the working condition judgment logic of this algorithm includes a four-layer structure: emergency working condition judgment, weighted distance matching degree judgment, frequency statistics judgment based on historical windows, and time consistency judgment. It should be noted that the emergency working condition judgment does not participate in the previous decision-making logic, that is, it is not evaluated through scoring, nor does it participate in the weight allocation process. It is directly and forcibly judged by the abnormal type feature lookup table and has the highest priority.

[0055] Operating condition determination formula: In the formula, As the score threshold, ; This formula means: when emergency features such as gas outburst or fire are detected, the corresponding emergency condition is directly determined; if no emergency condition is detected, then... The function scores the conditions and assigns a condition label to the corresponding working conditions; if no emergency condition is found and the output of the scoring function is below the scoring threshold, it indicates that an unknown working condition has been detected.

[0056] In one example of the present invention, step S30 includes the following steps: S31: Define decision variables: In the formula, For the first The set values ​​for each ventilation control device; among them, for the fan, , For the operating frequency, Hz; for the damper, , Percentage opening, %; For the total number of devices, In the formula, Where N is the total number of fans and N is the total number of dampers; S32: Design the overall objective optimization function F with the three optimization objectives of minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan; the expression of the overall objective optimization function F is: .

[0057] In one example of the present invention, in step S32, a general objective optimization function F is designed with three optimization objectives: minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan; specifically, it includes the following steps: S321: Construct the energy consumption minimization objective function: where, The expression for the wind turbine energy consumption model is: In the formula, For wind turbine Base power consumption under no-load conditions; The coefficient for the first-order term represents the linear losses of the wind turbine; The coefficient for the quadratic term represents the mechanical losses of the fan; The coefficient for the cubic term represents the aerodynamic power of the fan; For wind turbine Operating frequency; This model replaces the traditional one with a complete trinomial. The model incorporates mechanical losses and motor iron losses, resulting in superior simulation and more accurate results.

[0058] The expression for the damper energy consumption model is: In the formula, For air damper The drag coefficient, ; For air damper The degree of opening, ; The expression for the energy minimization objective function is: S322: Construct an objective function to maximize the uniformity of airflow distribution based on digital twins, where, The expression for the air volume prediction model based on digital twins is as follows: In the formula, Regarding the damper opening The impedance matrix; Regarding the power of the wind turbine Wind turbine power vector; The objective function for maximizing the uniformity of airflow distribution based on digital twins is expressed as follows: In the formula, The number of key roadways and working faces; For tunnels based on digital twin CFD models Predict air volume; This represents the average air volume for the entire mine. ; The goal is to minimize the relative variance between the air volume in each alley and the average air volume, to ensure balanced airflow distribution, and to avoid insufficient or excessive local ventilation.

[0059] S323: Construct the objective function for equipment life leveling, the expression of which is: In the formula, For wind turbine Rated power; respectively wind turbine Upper and lower limits of frequency; For air damper The rated opening degree; They are air dampers The upper and lower limits of the opening; This objective encourages equipment to operate near its rated operating conditions, avoiding lifespan reduction caused by prolonged operation outside of rated conditions, and achieving a balanced lifespan across multiple devices.

[0060] S324: Edge-side condition labeling The dynamic weight adjustment overall objective function F is expressed as: In the formula, These are the edge-side working condition labels. Dynamic weights.

[0061] Operating condition adaptive weighting mechanism: intelligently adjusts and optimizes target priorities based on real-time operating conditions to achieve automatic switching from "energy saving mode" to "safety mode".

[0062] Based on edge-side working condition labels Dynamic weight adjustment: In one example of the present invention, step S40 specifically includes the following steps: S41: Perform initialization settings, generate particle positions within feasible locations. That is, the particles are combinations of device set values, and a fitness evaluation is performed to match the overall objective optimization function. The expression for the fitness function is: In the formula, These correspond to the energy consumption component, the uniformity component, and the lifetime component, respectively, which are the three optimization objectives. The expression for the individual optimal position update rule is: In the formula, The particle number represents a different candidate solution, and each particle represents a possible ventilation strategy. The position vector parameter represents the particle's position, i.e., the particle's position vector. In the The position within a round corresponds to the value vector of the ventilation equipment; the spatial search dimension of this vector. This refers to the total number of fans and dampers, with each component corresponding to a set value for a single device. Component representation: If a mine has two ventilation fans and three ventilation doors, then: This indicates that the frequency of fan 1 is 42.5Hz, the frequency of fan 2 is 38.2Hz, the opening degree of damper 1 is 65%, the opening degree of damper 2 is 45%, and the opening degree of damper 3 is 80%.

[0063] Represents the individual's optimal vector parameters, storing particles. The optimal solution found during the search process; initial stage Iterative updates are performed; if the updated position is better, then it is updated to... Otherwise, keep it; S42: Initialization Each particle is randomly generated within a feasible region. This describes a device's configuration scheme: In the formula, Let be the vector space of the particles; This is the device lower bound vector; This is the upper limit vector for the device; it represents the minimum and maximum setpoints allowed for the device based on the device nameplate, safety operating procedures, and safety restrictions. The fitness function for each particle is obtained based on the optimization function in the multi-objective optimization decision module: Since there is more than one piece of equipment underground when applying to real-world mining scenarios, a core memory variable needs to be added. , is used to characterize the global historical optimal position; its expression is: In the formula, This is a weighted fitness function; The number of particles; This represents the current iteration number; S43: At the start of the algorithm, all particles are randomly initialized with their positions, their fitness is calculated, and the particle with the lowest fitness is found. And set the initial global optimal position. Its expression is: After each iteration from initialization, the particle position is updated, and the global optimal position variable is... It will compare the new positions generated by all particles, and if a better position is found, it will be updated. Otherwise, it remains unchanged; among them, The update mechanism expression is: The algorithm will search for all particles within the region, that is: S44: Particle velocity in the Weighted Sum Particle Swarm Optimization (WS-PSO) algorithm. This is a key search dynamics parameter that determines the search direction and step size of the particle in the solution space. Therefore, update rules for particles within a region can be formulated based on individual optimal position update rules and global optimal position update rules to characterize changes in device state. The expression for the particle update rule is: In the formula, This is the inertial term, meaning it maintains the search direction from the previous moment; This represents the individual experience term, meaning that the particle learns from the individual's historical best position, and the particle optimizes instructions based on the device's historical operating experience. This represents the social learning term, where each particle learns from the historical best position found by the entire population. Particles draw on the optimization experience of other devices or regions, even if a particle's own historical best position is not the best. Even if it's not ideal, you can still learn from the best experiences of the group. Obtain from; After the algorithm completes its iteration rounds, The final value is the global optimal solution: This represents the maximum number of iterations. Final algorithm output This optimized control command is then sent to the terminal fan and damper. The following is an example of the generated command: set up This indicates that the system has two fans and three dampers. Fan 1 command: 67.5Hz; Fan 2 command: 54.2Hz; Damper 1 command: 70%; Damper 2 command: 55%; Damper 3 command: 85%.

[0064] According to a second aspect of the present invention, a mine ventilation optimization system based on an edge computing architecture, such as... Figure 2 As shown, it includes: a perception execution layer, an edge computing layer, and a cloud platform layer, wherein, The sensing and execution layer includes an environmental sensing unit and an execution control unit. The environmental sensing unit is configured to sense gas concentration, wind speed, and temperature information within the roadway. The execution control unit is configured to precisely control the airflow within the roadway by adjusting the frequencies of fans and dampers. The environmental sensing unit is based on the architecture and algorithm design of three types of sensors most common underground and best characterizing ventilation conditions: intrinsically safe or explosion-proof gas concentration sensors, wind speed sensors, and temperature sensors. These sensors are deployed with a certain spatial density and topology to form a sensing network covering the entire mine. The execution control unit includes intelligent regulating dampers (adjustable opening), frequency-controlled main ventilators, and frequency-controlled local ventilators. These actuators receive instructions from the cloud computing layer to achieve precise airflow control.

[0065] The edge computing layer includes multiple edge computing nodes, each edge computing node including a data acquisition and preprocessing module and an intelligent analysis module. The data acquisition and preprocessing module is configured to acquire gas concentration information, wind speed information, and temperature information and perform normalization processing to form a normalized data set. The intelligent analysis module is configured to extract features from the normalized data set and perform time-series prediction, anomaly detection, and identification of the current operating condition based on the extracted features. The cloud platform layer is configured to receive edge-side data from abnormal edge computing nodes, perform simulations based on digital twins, construct a global ventilation network model, and establish a multi-objective optimization model with the goals of achieving safety standards, minimum energy consumption, most uniform airflow distribution, and minimum equipment loss; solving this optimization model yields the optimal control command. The system obtains the specific set values ​​of each fan and damper, and sends the optimal control command to the execution control unit to achieve ventilation optimization; the cloud platform layer sends the optimized control command to the corresponding command execution device to complete the real-time adjustment of ventilation parameters; the system continuously monitors sensor data and begins the next round of perception, analysis, and decision-making, forming a closed loop of "perception, analysis, optimization, and control".

[0066] This optimized system adds an edge computing layer to the traditional cloud-edge structure, enabling some core computing power to be deployed and dynamically scheduled at the edge. The edge layer and the cloud can dynamically collaborate. Under normal circumstances, the computing power stays only at the edge, and cloud computing power is called in case of an anomaly. This significantly saves transmission bandwidth and cloud platform computing resources, reduces overall energy consumption and operation and maintenance costs. At the same time, the edge-to-end uses a local area network instead of a wide area network in the traditional mode, which improves the robustness of the network structure.

[0067] The optimization system employs a lightweight multi-dimensional feature fusion and adaptive anomaly recognition algorithm at its edge, including real-time data preprocessing based on dynamic windows, multi-task feature extraction, weighted Manhattan distance anomaly detection, and operating condition recognition model. It comprehensively characterizes the ventilation system status through five-dimensional feature vectors and combines sliding window statistics and time consistency verification to achieve accurate identification of anomaly types and continuous and stable recognition of operating conditions. It has a rapid response capability, especially for emergency conditions such as gas outbursts and fires.

[0068] The optimization system is designed with a multi-objective optimization and operating condition adaptive strategy in the cloud. It proposes a multi-objective ventilation optimization model based on weighted and particle swarm optimization algorithms. It can dynamically adjust the optimization weights (energy consumption, uniformity, equipment life) according to the operating condition type reported from the edge, realize intelligent switching from "energy saving mode" to "safety priority mode", and improve the overall safety and economy of the system.

[0069] This optimization system, while ensuring mine ventilation safety, establishes a three-objective optimization function. Through optimized control strategies, it effectively reduces the energy consumption of fan operation, extends equipment life, and balances the distribution of underground air volume, thus achieving the dual goals of safe production and energy conservation.

[0070] The cloud platform layer is the command center of the entire system, possessing powerful computing capabilities. It is responsible for processing data uploaded from the edge, performing multi-objective optimization, issuing control commands, and handling deep learning and OTA updates.

[0071] Under normal circumstances, the underground environment is relatively safe and stable. Raw data collected by underground sensors is processed by data acquisition and preprocessing modules and lightweight artificial intelligence analysis modules deployed on edge servers. If the prediction results are normal and no abnormal trends are detected, the computing power is only used at the edge, without needing to call the cloud server, which greatly improves computing efficiency. First, compared to cloud servers, edge servers are deployed closer to the underground terminal devices, significantly reducing transmission latency. Second, edge servers use LAN access or wired transmission to connect to the terminals, rather than the wide area network communication of the traditional "cloud-edge" structure, which improves the stability of the transmission network. In addition, since multiple edge computing nodes may be deployed underground, when a single node fails, the terminals it manages can be connected in parallel to other nodes with redundant computing power, which improves the robustness of the network structure.

[0072] However, edge servers have limited computing power. For complex problems involving multiple factors and for further optimization and training of their own models, cloud computing power needs to be invoked.

[0073] Taking the downhole ventilation optimization involved in this invention as an example, the cloud server collects the abnormal breakpoint data returned by each edge node. That is, when an edge node detects an abnormal state, it returns all edge-side data at the same time segment to the cloud, specifically including the edge-side's defined timestamp, raw sensor data, time-series prediction results, abnormal state detection data, abnormality type retrieval, confidence level, and operating condition conclusions. Subsequently, the cloud performs simulation based on digital twin technology to construct a global ventilation network model, and establishes a multi-objective optimization model with the goals of achieving safety standards, minimum energy consumption, most uniform airflow distribution, and minimum equipment loss. Finally, the optimization model is solved to obtain the specific setpoints for each fan and damper, and control commands are sent to the terminal equipment to achieve ventilation optimization.

[0074] The cloud platform layer mainly deploys three core modules, corresponding to multi-objective optimization decision-making, optimization algorithm solving, and control command issuance, respectively. In the multi-objective optimization decision-making part, this invention proposes a working condition adaptive weighted optimization method, which dynamically adjusts the priority of optimization objectives based on the working condition type identified at the edge. In the optimization algorithm solving part, this invention proposes an improved weighted particle swarm optimization (WS-PSO) algorithm to achieve intelligent trade-offs in multi-objective conflicts, realize global collaborative optimization, and provide a progressive optimization strategy.

[0075] In one example of the present invention, the intelligent analysis module includes: The feature extraction module is configured to extract features from normalized data by solving a feature extraction formula, which is: In the formula, For feature vectors; For sliding statistics windows; This represents the mean value within the window. The standard deviation within the window; The time series prediction model is configured to predict the next H time steps based on current features; where the time series prediction formula is... for: In the formula, This is the trend decay constant; To predict the step size; The anomaly detection model is configured to detect anomalous data based on feature space distance. The working condition recognition model is configured to output a five-dimensional feature vector. As input to the working condition identification module, the output is a working condition category label. It identifies five working conditions—unknown, normal, maintenance, gas outburst, and fire emergency—by using distance calculation and rule-based judgment.

[0076] In one example of the present invention, the anomaly detection model includes: The spatial distance unit, configured for calculating the weighted Manhattan distance of data features, is expressed as follows: In the formula, These are the weight vectors, corresponding to the five dimensions of the feature vectors; For feature vectors; To indicate the number within the history window The average value of each feature, i.e., the sliding window The center (mean vector) of all eigenvectors within the vector is the first... Each component represents the normal center point of the system under the recent normal operating mode; The threshold setting unit, configured to adaptively set the threshold, has the following expression: In the formula, These are the mean and standard deviation of the historical feature distances, respectively; This is the threshold coefficient; Among them, the threshold coefficient The expression is: In the formula, The basic threshold coefficient; To adjust the amplitude coefficient; This is the sensitivity coefficient; This serves as a reference for the maximum level of fluctuation. To measure the fluctuation of the anomaly detection results themselves; if the distance value of recent anomaly detections... Violent fluctuations, The value will increase; if the distance value of recent anomaly detection... If the change is very small, then The value will decrease.

[0077] in, The expression is: The confidence level evaluation unit is configured to determine whether data features are abnormal based on confidence level, where the expression for confidence level is: Among them, the confidence level evaluation directly reflects the proportion of abnormal distance to the total distance, reflecting the relative degree of abnormality. hour, It is at the threshold boundary; when hour, , indicating a high degree of abnormality; when hour, This indicates a highly normal level.

[0078] In one example of the present invention, the working condition identification model includes: The standardization processing unit, configured to annotate data features, has the following expression: In the formula, For the first Historical statistical mean and standard deviation of each feature; To prevent constants with zero denominators, here ; The first working condition judgment unit is configured to judge gas outburst and fire conditions based on an anomaly type feature lookup table. The emergency judgment expression for gas outburst is as follows: The emergency fire assessment is expressed as follows: The second operating condition judgment unit is configured to judge unknown operating conditions, normal generation, and maintenance status based on an operating condition judgment formula, wherein the expression of the operating condition judgment formula is: In the formula, The criteria are based on weighted distance matching degree, frequency statistics, and time consistency. The score threshold; Specifically, when emergency features such as gas outburst or fire are detected, the corresponding emergency condition is directly identified; if no emergency condition is detected, then... The function scores the conditions and assigns a condition label to the corresponding working conditions; if no emergency condition is found and the output of the scoring function is below the scoring threshold, it indicates that an unknown working condition has been detected.

[0079] In one example of the present invention, the cloud platform layer includes: The multi-objective optimization decision module is configured to design a total objective optimization function with the objectives of minimizing energy consumption, maximizing the uniformity of air volume distribution, and balancing equipment lifespan. The weighted particle swarm optimization module is configured to use the weighted particle swarm optimization algorithm to optimize the objective function and obtain the best control scheme. The control command issuing module is configured to issue the acquired optimal control scheme to the execution control unit.

[0080] In one example of the present invention, the multi-objective optimization decision module includes: The decision variable definition unit, configured for defining decision variables, has the following expression: In the formula, For the first The set values ​​for each ventilation control device; among them, for the fan, (Operating frequency, Hz); For dampers, (Percentage opening, %) For the total number of devices, ; The optimization function construction unit is configured to design the overall objective optimization function F with three optimization objectives: minimizing energy consumption, maximizing airflow distribution uniformity, and balancing equipment lifespan. The expression for the overall objective optimization function F is: .

[0081] In one example of the present invention, the optimization function construction unit includes: The energy consumption objective function subunit is configured to construct the energy consumption minimization objective function: where, The expression for the wind turbine energy consumption model is: In the formula, For wind turbine Base power consumption under no-load conditions; The coefficient for the first-order term represents the linear losses of the wind turbine; The coefficient for the quadratic term represents the mechanical losses of the fan; The coefficient for the cubic term represents the aerodynamic power of the fan; For wind turbine Operating frequency; This model replaces the traditional one with a complete trinomial. The model incorporates mechanical losses and motor iron losses, resulting in superior simulation and more accurate results.

[0082] The expression for the damper energy consumption model is: In the formula, For air damper The drag coefficient, ; For air damper The degree of opening, ; The expression for the energy minimization objective function is: The airflow objective function subunit is configured to construct an objective function that maximizes the uniformity of airflow distribution based on digital twins, wherein... The expression for the air volume prediction model based on digital twins is as follows: In the formula, Regarding the damper opening The impedance matrix; Regarding the power of the wind turbine Wind turbine power vector; The objective function for maximizing the uniformity of airflow distribution based on digital twins is expressed as follows: In the formula, The number of key roadways and working faces; For tunnels based on digital twin CFD models Predict air volume; This represents the average air volume for the entire mine. ; The goal is to minimize the relative variance between the air volume in each alley and the average air volume, to ensure balanced airflow distribution, and to avoid insufficient or excessive local ventilation.

[0083] The lifetime objective function sub-unit is configured to construct the equipment lifetime equalization objective function, and its expression is: In the formula, For wind turbine Rated power; respectively wind turbine Upper and lower limits of frequency; For air damper The rated opening degree; They are air dampers The upper and lower limits of the opening; This objective encourages equipment to operate near its rated operating conditions, avoiding lifespan reduction caused by prolonged operation outside of rated conditions, and achieving a balanced lifespan across multiple devices.

[0084] The overall objective function sub-unit is configured to be based on edge-side conditions. The dynamic weight adjustment overall objective function F is expressed as: In the formula, They are respectively edge side working conditions Dynamic weights.

[0085] It should be noted that the mine ventilation optimization system based on edge computing architecture of the present invention can also perform any of the processes described in the previously described mine ventilation optimization method based on edge computing architecture, and the specific details are not repeated here.

[0086] Specific examples: Suppose a coal mining face area in a coal mine is equipped with the following sensors: gas concentration sensor: intrinsically safe, range 0-5%, installed in the return airway of the working face; wind speed sensor: explosion-proof, range 0-10m / s, installed in the middle of the roadway; temperature sensor: intrinsically safe, range 0-100℃, installed near the coal mining machine.

[0087] Edge computing nodes are deployed in the mining area substation to manage all sensors and actuators in the working face: 1 main ventilation fan (frequency converter controlled, frequency range 30-60Hz, currently 45Hz); 2 local ventilation fans (frequency converter controlled, frequency range 20-50Hz, currently 35Hz and 38Hz respectively); 3 air dampers (opening range 0-100%, currently 70%, 65%, and 80% respectively).

[0088] Case 1: No abnormalities were found in the edge layer analysis results (taking gas concentration as an example). Table 2 shows the sensor data acquisition table. Wherein, sampling frequency: Collect data at 10 time points (window size) ); Buffer (BUF): Stores the 20 most recent data samples; Sliding window : Retrieve the latest 10 data points.

[0089] Calculate the statistic: mean: Range: Standard deviation threshold: .

[0090] Anomaly detection: Momentary gas concentration Within the threshold range; no correction required. .

[0091] Normalization: Minimum value of the buffer: Maximum buffer size: Normalized values: .

[0092] Feature extraction: A five-dimensional feature vector was calculated using the latest 10 normalized gas data points, with windowed normalized mean: Instantaneous rate of change: Short-term volatility: Periodic coding: ; Change in compression ratio: ; Eigenvectors: .

[0093] Time series forecasting: Predicting the next 5 time steps: ; ; ; ; All are within the normal range.

[0094] Anomaly detection: Calculate the weighted Manhattan distance (weights) ); Historical Window Feature Center: ;distance: ; Historical distance mean and standard deviation: Adaptive threshold: ;determination: ; Operating condition identification: Feature vector standardization: Distance to various working conditions: Normal production distance: Gas outburst distance: Fire distance: Match degree: Historical window frequency: Time consistency test: Normal for 10 consecutive time points; Score: ;determination: In summary, the timing prediction results are normal; routine detection is normal; operating condition identification is normal; only edge layer computing power needs to be used to maintain the current ventilation status, and the parameters of the fan and damper remain unchanged.

[0095] Case 2: Abnormal results in edge layer analysis (with) (Taking a gas outburst as an example) Abnormal data scenarios: The gas concentration rose sharply over five consecutive time steps after the initial time. Table 3 is... Table of gas concentration increases over 5 consecutive time steps after a certain time point Preprocessing: Sliding window mean: Range: Threshold: Current value: .

[0096] Feature extraction: Normalization processing: ; Eigenvectors: .

[0097] Anomaly Detection: Weighted Manhattan Distance: Adaptive threshold: ;determination: Confidence level: Anomaly type identification (based on Table 2): Dimension 1 (short-term mean): Significant deviation from the normal range ( ); Dimension 2 (instantaneous rate of change): consistently positive with a sudden increasing trend; Dimension 3 (short-term volatility): significantly increased. Operating Condition Identification: Emergency Gas Outburst Detection: Direct judgment: .

[0098] In summary, the edge layer analysis confirms abnormal gas concentration characteristics, identifying it as an emergency gas outburst situation requiring global optimization using cloud computing power. The data uploaded to the cloud includes: raw sensor data, timestamps, time-series prediction results (continuous upward trend), anomaly detection data (confidence level 0.825), anomaly type (abnormal gas concentration), and situation conclusion (gas outburst).

[0099] The following process is the cloud-optimized computing part.

[0100] Multi-objective optimization model construction: Decision variables (total number of devices: 6): Adaptive weighting for operating conditions (gas outburst condition): (Safety first); Objective function: Constraints: Gas concentration The state of emergency has been eased to ; ; ; .

[0101] Solving with WS-PSO algorithm: Number of particles: Maximum number of iterations: Parameter settings: .

[0102] Table 4 shows the initial particle positions (5 particles are shown here). Energy consumption fitness calculation (taking particle 1 as an example): Main ventilation fan energy consumption: Energy consumption of local fan 1: Energy consumption of local fan 2: Energy consumption of dampers: Total energy consumption: .

[0103] Airflow uniformity fitness calculation (digital twin simulation): Assume based on digital twin simulation: number of roadways Air volume Average air volume: Uniformity: .

[0104] Equipment lifespan balance adaptability calculation: Rated value: Balance: .

[0105] Overall fitness: Iterative optimization (100 iterations until WS-PSO algorithm converges): Global optimal fitness: Final optimized instruction vector: .

[0106] Table 5 shows the optimal setting values ​​for the equipment. Digital twin simulation verification: Gas concentration: decreased from 2.35% to 1.25%, peak concentration 1.5%; Wind speed distribution: The wind speed at the working face increased from 1.8 m / s to 3.2 m / s, and the airflow uniformity index improved from 0.0052 to 0.0021; Energy consumption assessment: It increased from 205.95 kW to 245.60 kW. The principle of safety priority must be met, but the increase in energy consumption is acceptable.

[0107] System closed-loop verification (edge ​​layer monitoring): (Expected) After 10 minutes: Gas concentration drops to 0.85%; (Expected) After 30 minutes: The system stabilizes, and the gas concentration at all measuring points... 1.0%; Operating condition identification: from gas outburst (3) to normal production (1) Cloud computing power release: Optimization complete, computing power returns to standby state.

[0108] The foregoing description, with reference to preferred embodiments, details an exemplary implementation of the mine ventilation optimization method and system based on edge computing architecture proposed in this invention. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of this invention, and various combinations can be made to the various technical features and structures proposed in this invention without exceeding the protection scope of this invention, which is determined by the appended claims.

Claims

1. A mine ventilation optimization method based on edge computing architecture, characterized in that, Includes the following steps: S10: Deploy edge computing nodes and environmental sensing units at key locations underground, and correct and normalize the gas concentration, wind speed and temperature information sensed by the environmental sensing units. S20: Extract feature vectors from normalized data, calculate the weighted Manhattan distance between the feature vectors and the center of the window, match the feature vectors with the weighted distance of the preset working condition feature matrix, and combine historical window frequency statistics and time consistency judgment to output working condition labels. ; S30: Construct a digital twin ventilation network model with the optimization objectives of minimizing design energy consumption, maximizing airflow distribution uniformity, and balancing equipment lifespan, based on the output operating condition labels. Construct the overall objective function; S40: The weighted particle swarm optimization algorithm is used to solve the overall objective function and obtain the optimal control command. The commands are then sent to the underground ventilation fans and dampers, and the control instructions are dynamically adjusted based on the real-time gas concentration, wind speed, and temperature information obtained by the environmental sensing unit. .

2. The mine ventilation optimization system based on edge computing architecture according to claim 1, characterized in that, In step S10, the gas concentration, wind speed, and temperature information sensed by the environmental sensing unit are corrected and normalized, including the following steps: S11: Uses a fixed frequency Collect gas concentration, wind speed, and temperature data, and based on the previous... The mean of the data and range Dynamically calculate the current time. The anomaly detection threshold range, where the anomaly detection threshold is... formula: In the formula, This represents the size of the sliding window. For the front Moving average of the data; For the front The range of the data; This is the threshold coefficient; This is the range conversion factor; S12: When new sampling data Check the sampled data during input. Does it fall within the threshold? If within the upper and lower limits, Within this range, the data is considered normal; if the sampled data... Data outside this range is considered outlier and will be corrected using a correction formula, which is: S13: The corrected sampling data Normalization processing mapping to Interval, to eliminate the influence of dimensions and obtain Time-normalized output The formula for normalization is: In the formula, For a set of buffers; The minimum value in the buffer; This represents the maximum value in the buffer.

3. The mine ventilation optimization system based on edge computing architecture according to claim 1, characterized in that, Step S20 includes the following steps: S21: Feature extraction is performed on the normalized data by solving the feature extraction formula, which is: In the formula, For feature vectors; For sliding statistics windows; This represents the mean value within the window. The standard deviation within the window; S22: Predict the next H time steps based on current features; where the time series prediction formula is... for: In the formula, This is the trend decay constant; To predict the step size; for Time-normalized output data; S23: Detecting abnormal data based on feature spatial distance; S24: Output five-dimensional feature vector As input to the working condition identification module, the output is a working condition category label. It identifies five working conditions—unknown, normal, maintenance, gas outburst, and fire emergency—by using distance calculation and rule-based judgment.

4. The mine ventilation optimization system based on edge computing architecture according to claim 3, characterized in that, Step S23 includes the following steps: S231: Calculate the weighted Manhattan distance of data features Its expression is: In the formula, These are the weight vectors, corresponding to the five dimensions of the feature vectors; For feature vectors; To indicate the number within the history window The average of the features; S232: Adaptive threshold setting Its expression is: In the formula, These are the mean and standard deviation of the historical feature distances, respectively; This is the threshold coefficient; Among them, the threshold coefficient The expression is: In the formula, The basic threshold coefficient; To adjust the amplitude coefficient; This is the sensitivity coefficient; This serves as a reference for the maximum level of fluctuation. To measure the fluctuation value of the anomaly detection results themselves; in, The expression is: S233: Determine whether data features are abnormal based on confidence level, where confidence level... The expression is: Among them, the confidence level evaluation directly reflects the proportion of abnormal distance to the total distance, reflecting the relative degree of abnormality. hour, It is at the threshold boundary; when hour, , indicating a high degree of abnormality; when hour, This indicates a highly normal level.

5. The mine ventilation optimization system based on edge computing architecture according to claim 3, characterized in that, Step S24 includes the following steps: S241: Perform data feature annotation processing. The expression is: In the formula, For the first Historical statistical mean and standard deviation of each feature; To prevent constants with zero denominators, here ; S242: Based on the anomaly type feature comparison table, determine the gas outburst and fire conditions, including emergency determination for gas outbursts. The expression is: Fire emergency determination Expressed as: S243: Determine unknown working conditions, normal generation, and maintenance status based on the working condition determination formula, whereby the working condition label... The expression for the determination formula is: In the formula, The criteria are based on weighted distance matching degree, frequency statistics, and time consistency. The score threshold; S244: When emergency features such as gas outburst or fire are detected, the corresponding emergency condition is directly determined; if no emergency condition is detected, then... The function scores the conditions and assigns a condition label to the corresponding working conditions; if no emergency condition is found and the output of the scoring function is below the scoring threshold, it indicates that an unknown working condition has been detected.

6. The mine ventilation optimization method based on edge computing architecture according to claim 1, characterized in that, Step S30 includes the following steps: S31: Define decision variables : In the formula, For the first The set values ​​for each ventilation control device; among them, for the fan, , For operating frequency; for dampers, , Percentage opening; For the total number of devices, In the formula, Where N is the total number of fans and N is the total number of dampers; S32: Design the overall objective optimization function F with the three optimization objectives of minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan; the expression of the overall objective optimization function F is: 。 7. The mine ventilation optimization method based on edge computing architecture according to claim 6, characterized in that, In step S32, the overall objective optimization function F is designed with three optimization objectives: minimizing energy consumption, maximizing the uniformity of airflow distribution, and balancing equipment lifespan. Specifically, this includes the following steps: S321: Construct the energy consumption minimization objective function: where, Wind turbine energy consumption model The expression is: In the formula, For wind turbine Base power consumption under no-load conditions; The coefficient for the first-order term represents the linear losses of the wind turbine; The coefficient for the quadratic term represents the mechanical losses of the fan; The coefficient for the cubic term represents the aerodynamic power of the fan; For wind turbine Operating frequency; This model replaces the traditional one with a complete trinomial. The model includes mechanical losses and motor iron losses; Damper energy consumption model The expression is: In the formula, For air damper The drag coefficient, ; For air damper The degree of opening, ; Energy minimization objective function The expression is: S322: Construct an objective function to maximize the uniformity of airflow distribution based on digital twins, where, Digital twin-based air volume prediction model The expression is: In the formula, Regarding the damper opening The impedance matrix; Regarding the power of the wind turbine Wind turbine power vector; Objective function for maximizing airflow distribution uniformity based on digital twin Its expression is: In the formula, The number of key roadways and working faces; For tunnels based on digital twin CFD models Predict air volume; This represents the average air volume for the entire mine. ; S323: Constructing the objective function for equipment life balancing Its expression is: In the formula, For wind turbine Rated power; respectively wind turbine Upper and lower limits of frequency; For air damper The rated opening degree; They are air dampers The upper and lower limits of the opening; S324: Edge-side condition labeling The dynamic weight adjustment overall objective function F is expressed as: In the formula, These are the edge-side working condition labels. Dynamic weights.

8. The mine ventilation optimization method based on edge computing architecture according to claim 1, characterized in that, Step S40 specifically includes the following steps: S41: Perform initialization settings, generate particle positions within feasible locations. That is, the particles are combinations of device set values, and a fitness evaluation is performed to match the overall objective optimization function. The fitness function is... The expression is: In the formula, These correspond to the energy consumption component, the uniformity component, and the lifetime component, respectively, which are the three optimization objectives. Individual optimal position update rule The expression is: In the formula, The particle number represents a different candidate solution, and each particle represents a possible ventilation strategy. The position vector parameter represents the particle's position, i.e., the particle's position vector. In the The position within a cycle corresponds to the value vector of the ventilation equipment. Represents the individual's optimal vector parameters, storing particles. The optimal solution found during the search process; initial stage Iterative updates are performed; if the updated position is better, then it is updated to... Otherwise, keep it; S42: Initialization Each particle is randomly generated within a feasible region. This describes a device's configuration scheme: In the formula, Let be the vector space of the particles; This is the device lower bound vector; The upper limit vector of devices; The fitness function of each particle is obtained based on the optimization function in the multi-objective optimization decision module. The expression is: Since there is more than one piece of equipment underground when applying to real-world mining scenarios, a core memory variable needs to be added. Used to characterize the global historical best position ; Its expression is: In the formula, This is a weighted fitness function; The number of particles; This represents the current iteration number; S43: At the start of the algorithm, all particles are randomly initialized with their positions, their fitness is calculated, and the particle with the lowest fitness is found. And set the initial global optimal position. ; Its expression is: After each iteration from initialization, the particle position is updated, and the global optimal position variable is... It will compare the new positions generated by all particles, and if a better position is found, it will be updated. Otherwise, it remains unchanged; among them, The update mechanism expression is: The algorithm will search for all particles within the region, that is: S44: Based on the individual optimal position update rule and the global optimal position update rule, formulate the particle update rule within the region to characterize the change in device state; among which, the particle update rule The expression is: In the formula, This is the inertial term, meaning it maintains the search direction from the previous moment; This represents the individual experience term, meaning that the particle learns from the individual's historical best position, and the particle optimizes instructions based on the device's historical operating experience. This represents the social learning term, where each particle learns from the historical best position found by the entire population. Particles draw on the optimization experience of other devices or regions, even if a particle's own historical best position is not the best. Even if it's not ideal, you can still learn from the best experiences of the group. Obtain from; After the algorithm completes its iteration rounds, The final value is the global optimal solution: In the formula, This represents the maximum number of iterations. Final algorithm output The optimized control commands are then sent to the terminal fans and dampers.

9. A mine ventilation optimization system based on an edge computing architecture, characterized in that, include: The layers consist of a perception and execution layer, an edge computing layer, and a cloud platform layer. The sensing and execution layer includes an environmental sensing unit and an execution control unit. The environmental sensing unit is configured to sense gas concentration, wind speed, and temperature information within the tunnel. The execution control unit is configured to precisely control the airflow within the tunnel by adjusting the frequency of the fan and damper. The edge computing layer includes multiple edge computing nodes, each edge computing node including a data acquisition and preprocessing module and an intelligent analysis module. The data acquisition and preprocessing module is configured to acquire gas concentration information, wind speed information, and temperature information and perform normalization processing to form a normalized data set. The intelligent analysis module is configured to extract features from the normalized data set and perform time-series prediction, anomaly detection, and identification of the current operating condition based on the extracted features. The cloud platform layer is configured to receive edge-side data from abnormal edge computing nodes, perform simulations based on digital twins, construct a global ventilation network model, and establish a multi-objective optimization model with the goals of achieving safety standards, minimum energy consumption, most uniform airflow distribution, and minimum equipment loss; solving this optimization model yields the optimal control command. The specific set values ​​of each fan and damper are obtained, and the optimal control command is sent to the execution control unit to achieve ventilation optimization.

10. The mine ventilation optimization system based on edge computing architecture according to claim 9, characterized in that, The cloud platform layer includes: The multi-objective optimization decision module is configured to design a total objective optimization function with the objectives of minimizing energy consumption, maximizing the uniformity of air volume distribution, and balancing equipment lifespan. The weighted particle swarm optimization module is configured to use the weighted particle swarm optimization algorithm to optimize the objective function and obtain the best control scheme. The control command issuing module is configured to issue the acquired optimal control scheme to the execution control unit.

Images