Digital twin monitoring optimization method and system based on intelligent main ventilator

Abstract

The application discloses a digital twin monitoring optimization method and system based on an intelligent main ventilator, which comprises the following steps: S1, deploying a multi-modal sensor to collect multi-source ventilation data of the intelligent main ventilator; S2, based on the multi-source ventilation data, constructing a deep reinforcement learning model to be deployed on an edge computing node to output a ventilator control strategy; S3, constructing a ventilation network topology and a ventilator digital twin ontology model, importing environmental influence data through the ventilation network topology, and combining a graph neural network to perform multi-dimensional optimization on the ventilator digital twin ontology model, so as to obtain an improved ventilator digital twin model; and S4, simulating a system state after the ventilator control strategy is executed through the improved ventilator digital twin model, outputting a ventilation safety index, and performing optimization and adjustment according to the ventilation safety index, so that dynamic optimization of parameters such as a ventilator rotating speed, a blade angle and a ventilation path is realized through combination of the deep reinforcement learning model and the digital twin technology.

Description

Technical Field

[0001] This invention relates to the field of ventilation fan data monitoring technology, and in particular to a digital twin monitoring optimization method and system based on intelligent main ventilation fans. Background Technology

[0002] Currently, the main ventilation fan systems in mines, tunnels and other scenarios generally suffer from resource waste. Under the traditional constant speed operation mode, the fan power utilization rate is less than 60%, and the speed or damper opening is adjusted by manual experience. It is impossible to achieve precise control based on dynamic parameters such as gas concentration and roadway resistance, resulting in serious energy waste.

[0003] Existing monitoring systems often independently monitor single indicators such as gas concentration and wind pressure, without establishing a coupling relationship between the ventilator and the network topology. When gas levels exceed limits or equipment malfunctions, they cannot quickly locate the affected area and adjust ventilation strategies accordingly, resulting in a delayed safety response. Furthermore, traditional digital twin models only simulate the fan body or ventilation network separately, without using graph neural networks to achieve deep integration of mechanical, fluid, and electrical multi-physics fields with the network topology, leading to high simulation errors and difficulty in supporting the coordinated optimization of energy saving and safety. Therefore, this paper proposes a digital twin monitoring and optimization method and system based on an intelligent main ventilator. Summary of the Invention

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

[0005] A digital twin monitoring optimization method based on intelligent main ventilation fans includes:

[0006] S1: Deploy multimodal sensors to collect multi-source ventilation data from the intelligent main ventilator;

[0007] S2: Based on multi-source ventilation data, a deep reinforcement learning model is constructed and deployed on edge computing nodes to output ventilation fan control strategies;

[0008] S3: Construct a ventilation network topology and a digital twin ontology model of the ventilation fan. Import environmental impact data through the ventilation network topology and combine it with a graph neural network to optimize the digital twin ontology model of the ventilation fan in multiple dimensions to obtain an improved digital twin model of the ventilation fan.

[0009] The digital twin model of the ventilator includes mechanical characteristic mapping, fluid characteristic simulation, and electrical characteristic simulation;

[0010] S4: By improving the digital twin model of the ventilation fan, the system state after the ventilation fan control strategy is simulated, the ventilation safety index is output, and the optimization and adjustment are carried out based on the ventilation safety index;

[0011] The improved digital twin model of the ventilator obtains basic parameters through the mechanical characteristic mapping, fluid characteristic simulation, and electrical characteristic simulation of the ventilator digital twin ontology model. These parameters are then transmitted to the ventilation network topology and the node features are updated through a graph neural network. Finally, the updated features are fed back to the fan to generate an improved digital twin model of the ventilator that includes the interaction between the fan and the network, and outputs ventilation safety indicators.

[0012] The multi-source ventilation data includes the equipment's operating parameters N and the ventilation network parameters M;

[0013] The operating parameter N includes the rotational speed. ,power The ventilation network parameter M includes wind resistance. Air volume distribution Wind pressure .

[0014] The process of constructing the deep reinforcement learning model is as follows:

[0015] Construct a state space S, integrate multi-source data, and build a state vector. ;

[0016] Construct motion space A and define the controllable operations of the ventilation fan, including speed adjustment. , `max` represents the maximum rotational speed; blade angle adjustment... The ventilation path is switched to create space for movement;

[0017] Construct the reward function: ,in, , and All are preset weighting coefficients. Indicates energy-saving rewards, Indicates a safety reward. To ensure stable rewards;

[0018] A deep reinforcement learning model is trained using the Actor-Critic framework, combined with the PPO algorithm optimization strategy, and augmented using collected historical multimodal data. The sample size is divided into training, validation, and test sets in a ratio of 8:1:1. The training environment is deployed on edge computing nodes. When the loss of the validation set no longer decreases after 100 consecutive rounds, training is stopped, and the optimal model parameters are saved to obtain a fully trained deep reinforcement learning model.

[0019] The ventilation network topology construction process is as follows:

[0020] The tunnels, ventilation equipment, and monitoring points are abstracted as nodes and edges in a graph theory model, forming a visualized network topology. ,in:

[0021] Node V includes equipment nodes, roadway intersection nodes, nodes defining airflow direction, and gas monitoring point nodes;

[0022] The alleyway serves as an edge E connecting nodes, and the network connection relationships are recorded through an adjacency matrix.

[0023] The process of constructing the digital twin ontology model of the ventilator is as follows:

[0024] By coupling the data from mechanical property mapping, fluid property simulation, and electrical property simulation, a multiphysics coupling equation is established and integrated into a data framework using a graph structure to obtain a digital twin ontology model of the ventilator.

[0025] The process of importing environmental impact data into the ventilation network topology is as follows:

[0026] The environmental impact data includes gas concentration data A and tunnel geometry data B;

[0027] The gas concentration data A and the roadway geometry data B are imported into the topology of the ventilation network. Each node stores the gas concentration A at the current moment, and each edge stores the roadway geometry data B, thus completing the import of the environmental impact data of the ventilation network topology.

[0028] The process of using graph neural networks to perform multi-dimensional optimization of the digital twin ontology model of the ventilation fan is as follows:

[0029] A graph neural network is used to fuse the digital twin ontology model of the ventilation fan with the ventilation network topology that incorporates environmental impact data through a three-layer architecture;

[0030] The three-layer architecture includes a physical layer, a data layer, and an algorithm layer;

[0031] The physical layer establishes the coupling relationship between the digital twin ontology model of the ventilation fan and the ventilation network topology based on multiphysics equations. The data layer uses a graph structure to uniformly represent the equipment and network parameters, including the characteristics of ventilation fan nodes, roadway nodes, monitoring point nodes, and fan network connection edges. The algorithm layer uses a multi-layer graph neural network to realize cross-scale data interaction and joint simulation.

[0032] A digital twin monitoring and optimization system based on intelligent main ventilation fans includes:

[0033] Data acquisition module: Deploys multimodal sensors to collect multi-source ventilation data from the intelligent main ventilator;

[0034] Initial control module: Based on multi-source ventilation data, a deep reinforcement learning model is built and deployed on edge computing nodes to output ventilation fan control strategies;

[0035] Digital Twin Module: Construct a ventilation network topology and a digital twin ontology model of the ventilation fan. Import environmental impact data through the ventilation network topology and combine it with a graph neural network to optimize the digital twin ontology model of the ventilation fan in multiple dimensions to obtain an improved digital twin model of the ventilation fan.

[0036] Monitoring and optimization module: By improving the digital twin model of the ventilation fan, the system state after the ventilation fan control strategy is executed is simulated, ventilation safety indicators are output, and optimization and adjustment are made based on the ventilation safety indicators.

[0037] The present invention has the following beneficial effects:

[0038] In this invention, firstly, through training a large number of samples and optimizing the strategy using the PPO algorithm, the deep reinforcement learning model can adapt to different working conditions. Under conditions such as changes in roadway geometry or dynamic fluctuations in gas concentration, the ventilation fan control strategy can be automatically adjusted without manual intervention. The deep reinforcement learning model is deployed on edge computing nodes and combined with the real-time simulation capability of the improved digital twin model of the ventilation fan, it can realize the rapid processing of multi-source ventilation data and the real-time output of control strategies.

[0039] Secondly, a multi-dimensional safety indicator system is constructed, which covers aspects such as gas concentration, wind pressure fluctuation, equipment status, and energy consumption. This system can assess the safety status of the ventilation system in real time and trigger corresponding optimization and adjustment strategies when safety indicators exceed limits. Furthermore, it achieves deep integration of multi-physics fields and network topology. By leveraging the three-layer architecture of graph neural networks, the mechanical, fluid, and electrical characteristics of the digital twin model of the ventilation fan are deeply integrated with the ventilation network topology to form an improved digital twin model of the ventilation fan, enabling cross-scale data interaction and joint simulation.

[0040] Finally, by combining deep reinforcement learning models with digital twin technology and based on multi-source ventilation data output control strategies, dynamic optimization of parameters such as fan speed, blade angle, and ventilation path is achieved, realizing long-term benefits and significantly reducing the energy consumption of the ventilation system. Attached Figure Description

[0041] Figure 1 This diagram illustrates the steps of the digital twin monitoring and optimization method and system based on an intelligent main ventilation fan proposed in this invention.

[0042] Figure 2 This is a system block diagram of the digital twin monitoring optimization method and system based on intelligent main ventilation fan proposed in this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example 1: As Figure 1 As shown, the digital twin monitoring and optimization method and system based on intelligent main ventilation fans proposed in this invention include:

[0045] S1: Deploy multimodal sensors to collect multi-source ventilation data from the intelligent main ventilator;

[0046] Collect multi-source ventilation data from the intelligent main ventilator. The multi-source ventilation data includes the equipment's operating parameters N and the ventilation network parameters M.

[0047] Specifically, the operating parameter N includes the rotational speed. ,power , represented as The ventilation network parameter M includes air resistance. Air volume distribution and wind pressure , represented as These data serve as the input foundation for subsequent deep reinforcement learning models;

[0048] Data acquisition is achieved by deploying multimodal sensors (such as Hall effect speed sensors and catalytic combustion gas sensors) on the main ventilation fan, in the underground environment, and at key nodes of the ventilation network. The data is then transmitted in real time to the data processing center via a five-risk transmission network. Data preprocessing involves filtering and outlier detection to preprocess the initial multi-source ventilation data, unify dimensions, and remove outliers, thus providing a data foundation for building a deep reinforcement learning model.

[0049] S2: Based on multi-source ventilation data, a deep reinforcement learning model is constructed and deployed on edge computing nodes to output ventilation fan control strategies;

[0050] The process of building a deep reinforcement learning model is as follows:

[0051] Construct a state space S, integrate multi-source data, and build a state vector. The dimensions cover the entire operation of the ventilation system. Through data standardization, the influence of dimensions is eliminated, and different types of data can be collaboratively used in model training.

[0052] Construct motion space A and define the controllable operations of the ventilation fan, including speed adjustment. ( (max is the maximum speed) and blade angle adjustment. The ventilation path switching (binary variables represent path selection) creates a space for action, adapting to the complex control needs of the ventilation fan.

[0053] Construct a reward function, which is expressed as follows:

[0054] ,in, , and All are preset weighting coefficients:

[0055] Energy-saving rewards are negatively correlated with fan power and are expressed as... , The energy-saving weight coefficient is obtained through prior experience. The energy-saving reward directly incentivizes the model to reduce energy consumption, making energy saving the core orientation of strategy optimization.

[0056] Indicates safety rewards and sets airflow distribution thresholds. ,when hour, ;

[0057] when hour, , For safety reward coefficient, As a safety penalty coefficient, and All of these methods involve consulting equipment manuals to find and pre-set the air volume distribution standards for the ventilation fans. Through a reward and punishment mechanism, the model is forced to take into account the safety of air volume distribution during strategy optimization, so as to avoid sacrificing ventilation safety for energy saving.

[0058] To stabilize the reward, the wind pressure fluctuation and wind resistance standard deviation are measured. ,and Stable rewards Pick Otherwise take ,in, To stabilize the reward coefficient, To stabilize the penalty coefficient, and All of these were determined by consulting the equipment manual, which revealed the pre-set standards for the fan's pressure fluctuation and resistance standard deviation. For wind pressure fluctuations, The threshold for wind pressure fluctuation. For wind resistance standard deviation, The standard deviation threshold for wind resistance. The threshold for wind pressure fluctuation. The standard deviation threshold for wind resistance;

[0059] The reward function guides the model's learning direction. During training, the model explores different actions (such as adjusting rotation speed or changing paths) to obtain corresponding reward values, gradually optimizing the strategy. For example, an action that reduces power ( If the action is enhanced, it is strengthened; otherwise, it is suppressed.

[0060] The deep reinforcement learning model is trained using the Actor-Critic framework, and the optimization strategy is combined with the PPO algorithm. The specific process is as follows:

[0061] The Actor network (policy network) takes the state space S as input and outputs the action probability distribution. By learning the mapping between state and action, it generates a control policy.

[0062] The Critic network (value network) takes the state space S as input and outputs the state value to evaluate the expected benefit of performing an action in the current state, thus assisting the Actor network in optimizing the strategy.

[0063] Finally, the PPO algorithm is used to constrain the policy update magnitude to prevent the policy from updating too quickly and causing instability.

[0064] By utilizing collected historical multimodal data, augmentation techniques (such as time series shifting and scaling) can be applied to expand the dataset to... The sample size was divided into training set, validation set, and test set in a ratio of 8:1:1.

[0065] Deploy the training environment on edge computing nodes (such as industrial servers configured with GPU accelerator cards), use distributed training to accelerate model convergence, monitor the reward function value and policy loss value in real time during training, and stop training when the validation set loss no longer decreases for 100 consecutive rounds, and save the optimal model parameters.

[0066] The trained model is deployed on an edge computing node, receiving multimodal data preprocessed by S1 in real time, generating a state space S, and outputting control strategies (such as speed adjustment amount, path switching command, etc.) through forward propagation. The strategies are sent to the fan frequency converter and damper actuator to achieve intelligent control.

[0067] S3: Construct a ventilation network topology and a digital twin ontology model of the ventilation fan. Import environmental impact data through the ventilation network topology and combine it with a graph neural network to optimize the digital twin ontology model of the ventilation fan in multiple dimensions to obtain an improved digital twin model of the ventilation fan.

[0068] The process of constructing the ventilation network topology is as follows:

[0069] The tunnels, ventilation equipment, and monitoring points are abstracted as nodes and edges in a graph theory model, forming a visualized network topology. ,in:

[0070] Node V contains three types of entities: equipment nodes such as ventilators and dampers, which are used to store equipment rated parameters and spatial locations; roadway intersection nodes, which define the airflow direction; and gas monitoring point nodes, which are associated with real-time concentration data.

[0071] The alleyway acts as an edge E connecting nodes, and the network connection relationships are recorded through an adjacency matrix, for example:

[0072] In a mine ventilation system, the main ventilation fan is set as the core node, and the roadways of each mining area are connected as edges, forming a tree-like topology structure that visually presents the airflow path.

[0073] The process of constructing the digital twin ontology model of the ventilation fan is as follows:

[0074] The digital twin model of the ventilation fan includes mechanical characteristic mapping, fluid characteristic simulation, and electrical characteristic simulation, specifically:

[0075] Mechanical characteristic mapping takes into account the impeller dynamics. It establishes the equation of motion by using the moment of inertia, damping coefficient and driving torque to simulate impeller rotation and bearing vibration.

[0076] Taking a steel impeller as an example, the moment of inertia can be estimated by the square of the mass and the radius. When the driving torque changes, the model can calculate the corresponding change in the impeller angular velocity and then predict the bearing vibration amplitude. If the calculated radial acceleration exceeds 50 m / s², the system will trigger an abnormal warning, indicating that there may be a mechanical fault.

[0077] The fluid characteristic simulation employs computational fluid dynamics methods to mesh the internal flow field of the ventilator:

[0078] The impeller area has a grid accuracy of 0.05 meters, the volute area has a grid accuracy of 0.1 meters, and the total number of grids exceeds 2 million. Through fluid dynamics equations, the internal air pressure distribution and efficiency curve of the fan are obtained. Under rated operating conditions, the model can predict the impeller outlet air pressure to be 2800 Pa, and the error with the actual operating data is controlled within 5%.

[0079] Electrical characteristic simulation is based on the principle of equivalent circuit of motor. Based on parameters such as rated voltage of 380 volts and current of 120 amps, the internal resistance and inductance of the motor are deduced, and the input power and power factor of the motor are calculated in real time. For example, the power factor is about 0.85 when fully loaded. The input power can be obtained by multiplying the voltage, current and power factor, providing accurate data for energy consumption assessment.

[0080] By coupling the data from mechanical property mapping, fluid property simulation, and electrical property simulation, a multiphysics coupling equation is established and integrated into a data framework using a graph structure to obtain a digital twin ontology model of the ventilator.

[0081] The process of importing environmental impact data into the ventilation network topology is as follows:

[0082] Environmental impact data includes methane concentration data A and tunnel geometry data B;

[0083] The environmental impact data (tunnel geometry B, gas concentration data A) are imported into the ventilation network topology, which means converting the environmental parameters of the physical space into digital attributes of the topology model, giving the topology physical meaning, and enabling the digital twin model to simulate the operating logic of the real ventilation system.

[0084] Import the tunnel geometry data B from the environmental impact data into the topology of the ventilation network;

[0085] Specifically, the tunnel geometry data B includes the length L and cross-sectional dimensions (width c, height h) parameters. The equivalent diameter D of the tunnel (i.e., the diameter of the equivalent circular pipe) is calculated, taking a rectangular tunnel as an example:

[0086] If the width c = 4 meters, the height h = 3 meters, and the perimeter is 2(c + h) = 14 meters, then the cross-sectional area is 12 square meters. The equivalent diameter D can be obtained by importing the data. 3.43 meters, this value will serve as a key geometric property of the tunnel;

[0087] Import the gas concentration data A from the environmental impact data into the topology of the ventilation network;

[0088] An initial concentration value is set at the gas sensor deployment point, and the concentration diffusion process is calculated based on the air volume and roadway volume. For example, if a roadway has a volume of 1000 cubic meters, an upstream air volume of 50 cubic meters per second, and an initial concentration of 0.8%, the downstream concentration will be diluted to about 0.486% after 10 seconds due to airflow. This dynamic calculation ensures that the digital twin model can reflect the changes in gas distribution in real time, providing a basis for safety assessment.

[0089] Each node stores the gas concentration A at the current moment, and stores the roadway geometry data B (length L, cross-sectional dimensions (width c, height h), equivalent diameter D parameters) for each edge (roadway), thus completing the import of environmental impact data of the ventilation network topology;

[0090] The process of multi-dimensional optimization of the digital twin ontology model of the ventilation fan using graph neural networks is as follows:

[0091] A graph neural network (GNN) is used to deeply fuse the digital twin ontology model of the ventilation fan with the ventilation network topology based on fused environmental impact data, forming an improved digital twin model of the ventilation fan. The fusion process consists of a three-layer architecture:

[0092] Physical layer: Establish the coupling relationship between the digital twin ontology model of the ventilation fan and the ventilation network topology based on multiphysics equations;

[0093] Data layer: Uses a graph structure to uniformly represent device and network parameters;

[0094] Algorithm layer: Utilizing multi-layer graph neural networks to achieve cross-scale data interaction and co-simulation;

[0095] The physical layer implementation process is as follows:

[0096] When the fan speed is adjusted, the digital twin model of the fan calculates the new impeller outlet air pressure. and air volume This data is transmitted to the ventilation network topology, which then recalculates the air pressure in each tunnel of the entire ventilation network based on the received air pressure and air volume data. and air volume ;

[0097] Specifically, the entire fusion process is an iterative calculation process. The digital twin ontology model of the ventilation fan and the ventilation network topology continuously exchange data and recalculate their respective parameters until a stable system state is reached.

[0098] The data layer implementation process is as follows:

[0099] Based on graph neural networks, a graph structure is used to uniformly represent device and network parameters, including the characteristics of fan nodes. Features of roadway nodes Node characteristics of monitoring points And the fan - network connection edge ;

[0100] Specifically, the characteristics of ventilation fan nodes Represented as ,in:

[0101] ω represents the rotational speed of the fan impeller, which reflects how fast the fan rotates. The unit is radians per second. The magnitude of the angular velocity directly affects the air volume and air pressure output of the fan.

[0102] It's the power of the ventilation fan. This indicates the motor torque, which is closely related to the load and rotation state of the fan. Sufficient torque is necessary to ensure the normal operation of the fan.

[0103] r is the radius of the impeller, which determines the size of the impeller. The impeller radius has a significant impact on the air volume and air pressure generated by the fan. The larger the radius, the higher the performance of the fan.

[0104] Monitoring node characteristics Represented as ,in:

[0105] The methane concentration at the monitoring point must be controlled within a safe range, and is usually expressed as a volume percentage.

[0106] It is the wind pressure at the monitoring point. It refers to the air temperature at the monitoring point. Air temperature affects the density and viscosity of air, which in turn affects the airflow characteristics of the ventilation system.

[0107] Lane node characteristics Represented as ,in:

[0108] L is the length of the tunnel. h and are the width and height of the tunnel, respectively, and D is the equivalent diameter.

[0109] formula Obtain;

[0110] Indicates the wind resistance at the tunnel node, This indicates the airflow distribution at the roadway nodes;

[0111] The characteristics of the wind turbine-network connection edge are: ,in:

[0112] This refers to the outlet air pressure of the ventilator, which is the pressure at which the ventilator forces air into the ventilation network. The unit is Pascal. The magnitude of the outlet air pressure determines the ventilator's ability to overcome the resistance of the ventilation network and drive airflow.

[0113] It refers to the inlet air volume of the ventilation network, that is, the airflow entering the ventilation network. The inlet air volume is related to the outlet air volume of the fan and is also affected by the air resistance of the ventilation network.

[0114] By defining the graph structure based on graph neural networks, it is possible to comprehensively and accurately describe the various components of the ventilation system and their interrelationships.

[0115] The implementation process of the algorithm layer is as follows:

[0116] Based on GNN cross-scale fusion and co-simulation, message transmission from the wind turbine to the network is realized, including the wind turbine outlet pressure. The message is transmitted to the connected lane nodes, updating the lane air volume and aggregating the entrance air volume of all connected lanes. The corrected fan speed is expressed by the formula:

[0117] ;

[0118] in, This indicates the new fan speed. This is the original fan speed. The preset speed adjustment coefficient, This refers to the airflow at the entrance of the alleyway. Rated air volume, which is the airflow that the fan should provide under design operating conditions;

[0119] Specifically, the inlet air volume of the ventilation network It will affect the working status of the fan when the inlet air volume is less than the rated air volume. At this time, the fan speed needs to be adjusted to adapt to the needs of the ventilation network. In this way, the fan speed can be dynamically adjusted according to the actual air volume requirements of the ventilation network.

[0120] Finally, a multi-layer graph neural network was used to achieve feature fusion, resulting in a complete digital twin model of the ventilation fan.

[0121] Specifically, the first step is to set the rated angular velocity of the ventilator. and rated power (i.e., the ventilation power that the fan should provide under the design conditions), then establish a CFD grid with an accuracy of 0.05 meters, and import the data, including the geometric parameters of the roadway (such as length, width, height, etc.) and the initial gas concentration;

[0122] Basic parameters are obtained by mapping mechanical characteristics, simulating fluid characteristics, and simulating electrical characteristics of the digital twin ontology model of the ventilation fan. These parameters are then transferred to the ventilation network topology and updated through a graph neural network. Finally, the updated features are fed back to the fan to generate an improved digital twin model of the ventilation fan that includes the interaction between the fan and the network, and outputs ventilation safety indicators.

[0123] S4: By improving the digital twin model of the ventilation fan, the system state after the ventilation fan control strategy is simulated, the ventilation safety index is output, and the optimization and adjustment are carried out based on the ventilation safety index;

[0124] By improving the digital twin model of the ventilation fan, the output of the deep reinforcement learning model deployed on the edge computing node is used to formulate the ventilation fan control strategy, including the speed adjustment amount. Blade angle adjustment amount Δα or ventilation path switching command (binary vector);

[0125] Then, the ventilation control strategy is partially updated through the digital twin model of the ventilation fan, the CFD mesh boundary conditions (impeller speed, inlet pressure) are reset, and the input parameters in the ventilation control strategy (speed adjustment amount Δn, blade angle adjustment amount Δα or ventilation path switching command) are updated and recalculated according to the path switching command through the ventilation network topology.

[0126] The improved digital twin model of the ventilation fan outputs air safety indicators every 100ms. ,like Trigger strategy optimization:

[0127] Specifically, ventilation safety indicators include ,in:

[0128] This indicates the safety index for methane concentration; when the methane concentration exceeds 1%, The value is 1 if it is 1, otherwise it is 0.

[0129] This indicates a safety index related to wind pressure fluctuations and wind resistance standard deviation. When wind pressure fluctuations exceed 200 Pa or wind resistance standard deviations exceed 0.05, The value is 1 if it is 1, otherwise it is 0.

[0130] This indicates equipment safety indicators, such as when bearing acceleration exceeds 50 m / s² or wind pressure prediction error exceeds 5%. The value is 1 if it is 1, otherwise it is 0.

[0131] Energy-related safety indicators: When the power factor is less than 0.85, The value is 1 if it is 1, otherwise it is 0.

[0132] According to the values ​​of ventilation safety indicators, when When any one of the values ​​is greater than 1, the control strategy for the ventilation fan is optimized and adjusted by sending a Modbus-TCP message to the frequency converter and damper actuator based on prior experience or equipment manual.

[0133] For example, At that time, the optimization adjustment is to immediately implement full-volume ventilation (Δn=0.1nmax, all dampers are fully open).

[0134] Example 2: Figure 2 As shown, the digital twin monitoring and optimization system based on the intelligent main ventilation fan includes:

[0135] Data acquisition module: Deploys multimodal sensors to collect multi-source ventilation data from the intelligent main ventilator;

[0136] Initial control module: Based on multi-source ventilation data, a deep reinforcement learning model is built and deployed on edge computing nodes to output ventilation fan control strategies;

[0137] Digital Twin Module: Construct a ventilation network topology and a digital twin ontology model of the ventilation fan. Import environmental impact data through the ventilation network topology and combine it with a graph neural network to optimize the digital twin ontology model of the ventilation fan in multiple dimensions to obtain an improved digital twin model of the ventilation fan.

[0138] Monitoring and optimization module: By improving the digital twin model of the ventilation fan, the system state after the ventilation fan control strategy is executed is simulated, ventilation safety indicators are output, and optimization and adjustment are made based on the ventilation safety indicators.

[0139] In the application, several formulas are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and simulating the most recent real situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so they will not be elaborated here.

[0140] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for digital twin monitoring optimization based on intelligent main fan, characterized in that, The application relates to a method for constructing a deep reinforcement learning model for intelligent main ventilation fan control, and belongs to the technical field of ventilation safety control. The method comprises the following steps: S1, deploying a multi-modal sensor to collect multi-source ventilation data of an intelligent main ventilation fan; S2, based on the multi-source ventilation data, constructing a deep reinforcement learning model to be deployed on an edge computing node to output a ventilation fan control strategy; Constructing a state space S, integrating multi-source data, constructing a state vector ; Construct motion space A and define the controllable operations of the ventilation fan, including speed adjustment. , `max` represents the maximum rotational speed; blade angle adjustment... The ventilation path is switched to create space for movement; Constructing the reward function: wherein, , and are preset weight coefficients, represents the energy-saving reward, represents the safety reward, is the stability reward; The deep reinforcement learning model construction process comprises the following steps: S3, constructing a ventilation network topology and a ventilation fan digital twin ontology model, importing environmental influence data through the ventilation network topology, and combining a graph neural network to perform multidimensional optimization on the ventilation fan digital twin ontology model, so that an improved ventilation fan digital twin model is obtained; The ventilation fan digital twin ontology model comprises mechanical characteristic mapping, fluid characteristic simulation and electrical characteristic simulation; The process of combining the graph neural network to perform multidimensional optimization on the ventilation fan digital twin ontology model is as follows: The graph neural network is used to fuse the ventilation fan digital twin ontology model and the ventilation network topology with the fused environmental influence data through a three-layer architecture; The three-layer architecture comprises a physical layer, a data layer and an algorithm layer; The physical layer establishes a coupling relationship between the ventilation fan digital twin ontology model and the ventilation network topology based on a multi-physical field equation, the data layer uniformly represents device and network parameters through a graph structure, and the algorithm layer realizes cross-scale data interaction and joint simulation by using a multi-layer graph neural network; S4, simulating a system state after the ventilation fan control strategy is executed by using the improved ventilation fan digital twin model, outputting a ventilation safety index, and optimizing and adjusting according to the ventilation safety index; 2. The smart master fan based digital twin monitoring optimization method of claim 1, wherein, The improved ventilation fan digital twin model obtains basic parameters through mechanical characteristic mapping, fluid characteristic simulation and electrical characteristic simulation of the ventilation fan digital twin ontology model, the basic parameters are transmitted to the ventilation network topology and the node features are updated through the graph neural network, finally, the updated features are fed back to the fan, the improved ventilation fan digital twin model containing the interaction relationship between the fan and the network is generated, and the ventilation safety index is output. The operating parameters N include rotational speed , power , the ventilation network parameters M include air resistance , air volume distribution , air pressure .

3. The smart master fan based digital twin monitoring optimization method of claim 1, wherein, The multi-source ventilation data comprises operation parameters N of a device and ventilation network parameters M; The deep reinforcement learning model is trained by using an Actor-Critic framework, a PPO algorithm is used to optimize the strategy, historical multi-modal data is collected, and the data is expanded to a sample size by using data enhancement, the training set, the validation set and the test set are divided according to 8:1:1, the training environment is deployed on the edge computing node, when the loss of the validation set does not decrease continuously for 100 rounds, the training is stopped, the optimal model parameters are saved, and a complete deep reinforcement learning model is obtained.

4. The smart master fan based digital twin monitoring optimization method of claim 1, wherein, The deep reinforcement learning model construction process further comprises the following steps: The roadway, ventilation equipment and monitoring points are abstracted as nodes and edges in a graph theory model to form a visual network topology, wherein: The ventilation network topology construction process is as follows: A node V comprises a device node, a roadway intersection node, a gas monitoring point node and a defined air flow direction; 5. The smart master fan based digital twin monitoring optimization method of claim 1, wherein, A roadway is an edge E connecting the nodes, and the network connection relationship is recorded through an adjacency matrix. The ventilation fan digital twin ontology model construction process is as follows:

6. The smart master fan based digital twin monitoring optimization method of claim 1, wherein, The data of mechanical characteristic mapping, fluid characteristic simulation and electrical characteristic simulation are coupled, a multi-physical field coupling equation is established, and the data are fused and uniformly represented and integrated into a data framework to obtain the ventilation fan digital twin ontology model. The process of importing environmental influence data into the ventilation network topology is as follows: The environmental influence data comprises gas concentration data A and roadway geometric size data B; 7. A digital twin monitoring optimization system based on intelligent main fan, according to the method of any one of claims 1-6, characterized in that, The gas concentration data A and the roadway geometric size data B are imported into the topology structure of the ventilation network, each node stores current-time gas concentration A, and each edge stores roadway geometric size data B, so that the ventilation network topology imports the environmental influence data. The application further discloses a ventilation safety control system. The ventilation safety control system comprises the following modules: a data acquisition module, which is configured to deploy a multi-modal sensor to collect multi-source ventilation data of an intelligent main ventilation fan; a ventilation network topology module, which is configured to construct a ventilation network topology based on the multi-source ventilation data; a ventilation fan digital twin ontology model module, which is configured to construct a ventilation fan digital twin ontology model based on the ventilation network topology and the multi-source ventilation data; a deep reinforcement learning model module, which is configured to construct a deep reinforcement learning model based on the ventilation network topology and the ventilation fan digital twin ontology model, and output a ventilation fan control strategy; a ventilation safety index output module, which is configured to simulate a system state after the ventilation fan control strategy is executed by using the improved ventilation fan digital twin model, output a ventilation safety index, and optimize and adjust according to the ventilation safety index. Initial regulation module: based on multi-source ventilation data, a deep reinforcement learning model is constructed and deployed on the edge computing node to output the ventilation fan regulation strategy; Digital twin module: a ventilation network topology and ventilation fan digital twin ontology model are constructed, environmental impact data are imported through the ventilation network topology, and a multi-dimensional optimization is performed on the ventilation fan digital twin ontology model by combining a graph neural network to obtain an improved ventilation fan digital twin model; Monitoring optimization module: the system state after the execution of the ventilation fan regulation strategy is simulated by the improved ventilation fan digital twin model, and ventilation safety indicators are output, and optimization and adjustment are performed according to the ventilation safety indicators.

Images