Digital twin driven adaptive control method for gas monitoring network
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA UNIV OF PETROLEUM (EAST CHINA)
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-26
AI Technical Summary
The problem of monitoring blind spots caused by sparse sensor deployment in confined spaces is difficult to achieve with existing technologies. This results in the inability to detect dangerous gas leaks in a timely manner, posing a safety hazard. Furthermore, increasing the number of sensors deployed would increase costs and is difficult to achieve in complex spaces.
By constructing a gridded mapping between physical space and digital twin space, and using measured data from sparse sensors to drive a virtual fluid dynamics model, a physical reconstruction from discrete point data to a continuous field is achieved. Combined with the digital twin's advanced simulation of gas diffusion paths, the network topology and wake-up strategy are dynamically adjusted to construct an adaptive control method.
It achieves improved monitoring coverage accuracy and data reliability without increasing the number of sensors, reduces deployment and maintenance costs, adapts to complex and confined space scenarios, enhances security and prevention capabilities, and improves the real-time performance and applicability of monitoring.
Smart Images
Figure CN122293673A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of confined space safety monitoring technology, specifically to a digital twin-driven adaptive control method for gas monitoring networks. Background Technology
[0002] Currently, routine monitoring of confined spaces primarily relies on sensor deployment for data collection. However, due to the complex structure of confined spaces, deployment costs, and equipment maintenance difficulties, sensor deployment generally suffers from sparsity, making it difficult to achieve dense coverage of the entire space and resulting in widespread monitoring blind spots. Within these blind spots, potential hazards such as dangerous gas leaks and abnormal concentrations cannot be detected in a timely manner, easily creating safety vulnerabilities and posing a serious threat to the safe operation of confined spaces.
[0003] Existing technologies often employ the method of increasing the number of sensors deployed, but this approach significantly increases the costs of equipment purchase, installation, and subsequent maintenance. Furthermore, it is difficult to achieve dense deployment in some complex and confined spaces, resulting in poor applicability. Other technologies use simple data interpolation methods to fill monitoring blind spots, but these methods do not consider the diffusion patterns of fluids within confined spaces, leading to large interpolation errors. They fail to reflect the true distribution of gas concentrations within the space and are therefore unsuitable as an effective basis for subsequent control decisions.
[0004] The rise of digital twin technology has provided a new technological path for confined space monitoring. Its core lies in constructing a precise mapping between physical space and digital twin space, realizing the digital replication and dynamic evolution of physical scenes. However, most existing monitoring technologies based on digital twins can only achieve a simple digital representation of physical space, failing to effectively utilize limited measured sensor data to drive the optimization of virtual models. They cannot achieve precise physical reconstruction from "discrete point" measured data to "continuous field" full-domain data, resulting in insufficient consistency between digital twin space and physical space. This makes it difficult to provide accurate spatiotemporal state benchmarks, and consequently, cannot support subsequent core needs such as on-demand control and risk warning. Summary of the Invention
[0005] This invention provides a digital twin-driven adaptive control method for gas monitoring networks to solve the monitoring blind zone problem caused by sparse sensor deployment in confined spaces under normal monitoring conditions in the prior art.
[0006] To address the above problems, this invention employs the following technical solution: a digital twin-driven adaptive control method for gas monitoring networks, comprising the following steps: S1. Construct a gridded mapping relationship between physical space and digital twin space, use sparse sensor measured data to drive a virtual fluid dynamics model, realize the physical reconstruction from discrete point data to continuous gas field, and obtain the global spatiotemporal state benchmark and benchmark flow field; S2. Based on the dynamic reconstruction results of the gas field, the high-risk area mask and the reference flow field parameters, the gas diffusion path and arrival time are predicted in advance by digital twin, and the dormant sensor nodes are accurately awakened. S3. Map gas concentration risk to communication link cost, and dynamically adjust network topology, relay nodes and link weights by combining wake-up node set to build a low-latency, high-redundancy adaptive physical network; S4. Establish a two-way feedback mechanism between the physical perception layer and the digital twin decision layer, feed back the physical network status to the digital twin model for parameter correction, and optimize the wake-up strategy and network parameters in reverse through the twin inference results to achieve system adaptive updates and fault self-healing.
[0007] Furthermore, S1 includes S1.1 defining the physical domain of the confined space. Divided into Each control volume is a non-overlapping control volume, and all control volumes together constitute a discrete mesh set. Define a set of geometric properties for any control volume. With the flow field state vector : ; In the formula, For the first The geometric center coordinates of each grid. For the first The physical volume of each grid; ; In the formula, For the first The reference wind speed vector within each grid. air density, For the first The turbulence intensity of each grid, The effective diffusion coefficient; The reference flow field was obtained by solving the Navier-Stokes equations and coupled with... Turbulence model to calculate turbulence intensity and effective diffusion coefficient : ; ; ; ; In the formula, For Hamiltonian operators, air density, This is the reference wind speed vector within the grid. For turbulent kinetic energy, For turbulent viscosity, For turbulent kinetic energy, Prandtl number, The turbulent kinetic energy term generated by the average velocity gradient. The turbulent kinetic energy dissipation rate, For the first Turbulent kinetic energy of each grid For the first Reference wind speed vector within each grid The model, The molecular diffusion coefficient is... For turbulent Schmidt number; Each grid , and By integrating the data, a baseline flow field database containing turbulence information is obtained. ; S1.2 settings At any given moment, all nodes in the physical mesh sensor network are traversed, and sensor nodes that are currently in working condition, collecting and uploading data normally are selected to form an active node set. For any node Define any node in State vector at time step : ; In the formula, For nodes exist The gas concentration collected at all times, For nodes exist The rate of change of gas concentration at time t; For nodes exist The original sampled signal at time t is subjected to improved Kalman filtering, and the filtering process follows the formula: ; ; ; In the formula, For nodes exist The predicted state value at time t, i.e., based on The state at time is derived The estimated state value at time step and the state vector Consistent in dimensions, including predicted gas concentrations and rates of change of concentration. Here is the state transition matrix. For nodes exist The state vector at time t, for Time-based process noise, For nodes exist The sampled value of the original gas concentration with noise at a given time. For the observation matrix, The true state vector contains Real gas concentration at any given time and the actual concentration change rate , for Observation noise at any given moment For nodes exist The posterior estimate of the state at time t. Kalman gain for signal filtering; Output node is processed by filtering. High confidence concentration value after noise reduction ; High confidence concentration value after noise reduction According to the discrete grid set corresponding to the sensor node Integrating to form a sparse concentration vector : ; In the formula, To activate the node exist At time 1, the concentration estimate after Kalman filtering. This represents the total number of activated nodes.
[0008] Furthermore, S1 includes S1.3, which is the sparse concentration vector output from S1.2. Inverse solution of gas leak source strength and location Construct a weighted cost function : ; ; ; In the formula, For nodes The confidence weight coefficient, As the turbulence penalty factor, For nodes At discrete grid locations Turbulence intensity at that location This is a modified Gaussian plume model for confined spaces. For digital twin models at arbitrary discrete grid locations within a confined space The predicted gas concentration value at the location, Indicating the strength of the gas leak source. Let be the gas diffusion coefficient. Discrete grid position Location of the leak source distance, For digital twin models at nodes Location The predicted gas concentration value at the location, The regularization coefficient is . Intensity of gas leak source The square of the L2 norm; By solving the weighted cost function The minimum value is used to estimate the location of the leak source. and leakage intensity ; S1.4 Leakage source location estimated based on S1.3 and leakage intensity Construct source term function : ; The global concentration field is then reconstructed using unsteady convection-diffusion equations: ; In the formula, For the first Gas concentration at each time step For the first Gas concentration at each time step For digital twin simulation time step, for Gas concentration distribution at time gradient vector, For gradient operators, For the first The spatial distribution gradient of gas concentration at each time step. To estimate the gas source term; by Using boundary constraints, we obtain all discrete meshes within the confined space. exist Gas concentration value at time By integrating all grid concentration values in grid order, a global gas concentration distribution matrix is obtained. : ; In the formula, For the first Discrete grid cells exist Gas at a given moment.
[0009] Furthermore, S1 includes S1.5 constructing a gradient prediction operator that incorporates the time derivative. The concentration change rate was obtained by finite difference calculation: ; ; ; In the formula, The rate of change of concentration over time. For in position , The gas concentration at time , For in position , The gas concentration at time , For in position , The danger determination function at time t, For predicting time coefficients, The dangerous concentration threshold, Masking for high-risk areas, For the first Each grid in The danger threshold at any given moment.
[0010] Furthermore, S2 includes S2.1 constructing a discrete mesh set. With sensor node set Static mapping relationship matrix between : ; In the formula, For the first Pre-defined deployment coordinates of each sensor node; During real-time operation, the high-risk area mask obtained based on S1.5 is used. Generate a preliminary candidate wake-up set : ; In the formula, This is the initial candidate wake-up set. For the first One sensor node, For the existence of masks belonging to high-risk areas The One grid; right Perform neighborhood expansion to obtain the expanded candidate wake-up set. : ; In the formula, To expand the candidate wake-up set, For the first One sensor node, For sensor nodes With sensor nodes Spatial distance between them This is the safety buffer distance threshold; S2.2 For the expanded candidate wake-up set Any sensor node in Calculate the gas front arrival node Expected time Introducing local transport velocity : ; In the formula, For position Local transport velocity at the location, For position The reference wind speed vector at that location, This is the transport time scale coefficient. For position The effective diffusion coefficient at that location, The length of the grid feature; From the location of the leak source To sensor node Location Shortest fluid trace Integrate to obtain the estimated arrival time of the gas front. : ; In the formula, For the current moment, To the location of the leak source To sensor node Location The shortest fluid trace; according to Configure sensor nodes wake-up command time : ; In the formula, For sensor warm-up time, This represents network latency margin.
[0011] Furthermore, S2 includes S2.3 constructing sensor nodes. Activation utility function : ; In the formula, , , , For the utility function weights, For nodes Location exist Predicted gas concentration at any given time The time scale constant, For sensor nodes The remaining battery power, Fully charge the sensor nodes. This is the set of sensor nodes that are currently in operation. For sensor nodes With sensor nodes Spatial distance, It is a smoothing factor; Expanded candidate wake-up set Select a subset of wake-up nodes from the list to construct the final wake-up list. : ; In the formula, To expand the candidate wake-up set any subset, For subset The sum of the activation utility values of all sensor nodes in the system; And satisfy channel capacity constraints: ; In the formula, The preset channel capacity limit; S2.4 gateway To the final wake-up list The node in the middle sends a wake-up frame: ; In the formula, For Bloom filter mask, This is the relative time offset; node The operating status is determined based on the signal strength and the Bloom filter verification result. ; In the formula, For sensor nodes The next work status, The preset signal strength threshold; In In this state, the node sets a wake-up timer based on the relative time offset and enters high-frequency sampling mode.
[0012] Furthermore, S3 includes S3.1 constructing a weighted directed graph of the sensor network, for any two sensor nodes... , Define link transmission cost : ; In the formula, For risk aversion coefficient, For sensor nodes To sensor node Historical package delivery rate, For sensor nodes With sensor nodes Gas risk concentration in inter-linkage This is the critical risk threshold for gases. Energy weighting coefficient, This represents the initial charge level of the node. For nodes Remaining battery power; Risk concentration on the link path pass obtained through linear sampling: ; In the formula, For sensor nodes With sensor nodes The straight path between them In order to be in At any given moment, the global gas concentration field In the middle, spatial location Corresponding predicted gas concentration; Construct a weighted adjacency matrix for the entire network based on signal strength thresholds. : ; In the formula, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between For sensor nodes To sensor node The received signal strength, This is the receiver sensitivity threshold; when When this occurs, it is considered that the physical link is unreachable, and communication is interrupted; S3.2 Based on the sensor node set The set of dormant nodes is obtained. : ; from Selecting a set of relay nodes This minimizes the objective function: ; In the formula, The wake-up cost weighting coefficient, The fixed energy cost of waking up a relay node, This is the energy balance weighting coefficient. For environmental risk weighting coefficients, For relay nodes The remaining battery power, For relay nodes Location exist Predicted gas concentration at any given time This refers to the safe concentration threshold for the gas. active node set for: ; Depend on Constitutes the final active topology : ; S3.3 Define Gateway Zero potential energy point, any sensor node The potential energy is determined by the cumulative link cost, and the calculation formula is: ; In the formula, For sensor nodes The potential energy value, For sensor nodes The set of all neighboring sensor nodes, For sensor nodes The potential energy value, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between; The next hop node is determined based on the potential energy of the data flow direction. The formula for calculating the set of next hop nodes is as follows: ; In the formula, For sensor nodes The set of next-hop nodes; Output the routing table for each source node. .
[0013] Furthermore, S3 includes S3.4 defining sensor nodes. Dynamic latency requirements Dynamic latency requirements It is linked to the gas diffusion rate, and the calculation formula is: ; In the formula, Based on latency, The concentration change rate weighting coefficient, For sensor nodes Location The rate of change of concentration at that location; Define business priorities : ; In the formula, This is the priority weight coefficient. For nodes The actual concentration at the location The maximum gas concentration threshold; Based on the business priority of each node The sorting results are used to allocate dedicated guaranteed time slots to the highest priority nodes and ordinary transmission time slots to the remaining nodes. The time slot allocation results of all nodes are then compiled into a unified transmission time slot table. , For business priorities The highest-level node is allocated a dedicated guaranteed time slot, based on the receiver. Feedback adjustment node transmit power, the power adjustment formula is: ; In the formula, The transmission power at the previous moment. This is the power adjustment coefficient. For the target received signal strength, The received signal strength of the previous frame. For risk compensation power; Output the transmit power setting value for each sensor node. and transmission time slot table .
[0014] Furthermore, S4 includes S4.1, which involves assessing the remaining energy of each sensor node. Delivery rate of each link Length of data queue for each sensor node ,definition The state vector of the entire network at time step : ; In the formula, For the first The remaining energy of each sensor node, For the first Delivery rate of each link, For the first The length of the data queue for each sensor node; On the digital twin server side, through Norm minimization reconstructs the original state: ; Output the reconstructed physical state of the entire network ; S4.2 Define the model parameter vector to be corrected. : ; In the formula, For boundary wind speed, The turbulent diffusion coefficient; pass The model is evolved using a set of samples, and the evolution formula is as follows: ; In the formula, For the first Predicted parameters for a set of samples, These are the posterior analysis parameters from the previous time step. Evolutionary noise; Using the output of S1.2 Adjust parameters and introduce As an observation operator, the continuous field is mapped to the discrete sensor position, and the update formula is: ; In the formula, For the first Posterior analysis parameters of a set of samples, To assimilate the Kalman gain for the data, For solving convection-diffusion equations; right Posterior analysis parameters of a set of samples By taking the arithmetic mean and eliminating the influence of noise, the corrected optimal model parameters are obtained. .
[0015] Furthermore, S4 includes S4.3 with a global concentration field. The reconstructed physical state of the entire network System risk level Define the system state space as input. With action space : ; ; In the formula, For the global concentration field, This represents the average physical state of the entire network. , , For utility function weight adjustment, Adjustments made to mitigate risk. To avoid route adjustments, Adjustment for latency weights; The PPO algorithm is used to train the policy network. Output the optimal weight configuration: ; In the formula, This is the set of hyperparameters for the algorithm at the next time step. The set of basic hyperparameters; S4.4 Defines System Operating Modes Based on the system state, adaptive mode switching is achieved, and the state transition equation is: ; In the formula, This is the highest concentration in the entire venue. This is the danger alarm threshold. This represents the link failure rate. The threshold for the link failure rate. The minimum remaining energy for all nodes in the network. The energy threshold for node death; against The mode is re-triggered in S3.2, and failed nodes are forcibly removed, resulting in an updated set of dormant nodes. : ; In the formula, For the updated set of dormant nodes, For a set of dormant nodes, This is the set difference operator. A subset of failed nodes; Output system current operating mode and system timing control parameters.
[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a mapping between physical space and digital twin space, using limited measured data to drive a virtual fluid dynamics model, achieving precise reconstruction from "discrete points" to a "continuous field." This eliminates the need to increase the number of sensors, reducing deployment and maintenance costs while improving monitoring coverage accuracy and data reliability. Simultaneously, the reconstructed continuous field data provides a precise spatiotemporal state benchmark, supporting subsequent on-demand operations such as routing scheduling and dynamic latency control. This invention balances economy and applicability, adapting to various complex and confined space scenarios. By optimizing data transmission and dynamic control, it improves monitoring real-time performance and strengthens security and control capabilities. With a complete technical chain and strong practicality, it possesses high promotional value. Attached Figure Description
[0017] Figure 1 This is a flowchart of the technology of the present invention. Detailed Implementation
[0018] The present invention will be further illustrated below with reference to embodiments. These embodiments are for illustrative purposes only and are not intended to limit the invention in any way. It should be understood that the described embodiments are merely some, not all, of the embodiments described in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.
[0019] like Figure 1 As shown, a digital twin-driven adaptive control method for a gas monitoring network includes the following steps: S1.1 defines the physical domain of the confined space. Divided into Each control volume is a non-overlapping control volume, and all control volumes together constitute a discrete mesh set. Define a set of geometric properties for any control volume. With the flow field state vector : ; In the formula, For the first The geometric center coordinates of each grid. For the first The physical volume of each grid; ; In the formula, For the first The reference wind speed vector within each grid. air density, For the first The turbulence intensity of each grid, The effective diffusion coefficient; The reference flow field is obtained by solving the Navier-Stokes equations and coupled with... Turbulence model to calculate turbulence intensity and effective diffusion coefficient : ; ; ; ; In the formula, For Hamiltonian operators, air density, This is the reference wind speed vector within the grid. For turbulent kinetic energy, For turbulent viscosity, For turbulent kinetic energy, Prandtl number, The turbulent kinetic energy term generated by the average velocity gradient. The turbulent kinetic energy dissipation rate, For the first Turbulent kinetic energy of each grid For the first Reference wind speed vector within each grid The model, The molecular diffusion coefficient is... For turbulent Schmidt number; Each grid , and By integrating the data, a baseline flow field database containing turbulence information is obtained. .
[0020] This step establishes a full-parameter mapping from physical space to digital space, achieves spatial discretization, and also calculates by introducing a turbulence model. and This lays the foundation for subsequent weight calculations and diffusion models.
[0021] S1.2 settings At any given moment, all nodes in the physical mesh sensor network are traversed, and sensor nodes that are currently in working condition, collecting and uploading data normally are selected to form an active node set. For any node Define any node in State vector at time step : ; In the formula, For nodes exist The gas concentration collected at all times, For nodes exist The rate of change of gas concentration at time t; For nodes exist The original sampled signal at time t is subjected to improved Kalman filtering, and the filtering process follows the formula: ; ; ; In the formula, For nodes exist The predicted state value at time t, i.e., based on The state at time is derived The estimated state value at time step and the state vector Consistent in dimensions, including predicted gas concentrations and rates of change of concentration. State transition matrix , used to The node state transitions at time [time]. At that moment, among them The sampling time interval of the sensor is used to suppress noise by utilizing the physical inertial constraint that the gas concentration does not change abruptly. For nodes exist The state vector at time t, for Time-based process noise, For nodes exist The sampled value of the original gas concentration with noise at a given time. Observation matrix This indicates that only the gas concentration was observed. The true state vector contains Real gas concentration at any given time and the actual concentration change rate , for Observation noise at any given moment For nodes exist The posterior estimate of the state at time t. Kalman gain for signal filtering; Output node is processed by filtering. High confidence concentration value after noise reduction ; High confidence concentration value after noise reduction According to the discrete grid set corresponding to the sensor node Integrating to form a sparse concentration vector : ; In the formula, To activate the node exist At time 1, the concentration estimate after Kalman filtering. This represents the total number of activated nodes.
[0022] This step cleans the raw data uploaded from the physical layer, eliminating common electromagnetic interference and sensor drift in confined spaces, ensuring that the boundary conditions input to the digital twin model are accurate true values, and preventing divergence in subsequent inversion calculations due to single-point false alarms.
[0023] S1.3 Sparse concentration vector based on the output of S1.2 Inverse solution of gas leak source strength and location Construct a weighted cost function : ; ; ; In the formula, For nodes The confidence weight coefficient, As the turbulence penalty factor, For nodes At discrete grid locations Turbulence intensity at that location This is a modified Gaussian plume model for confined spaces. For digital twin models at arbitrary discrete grid locations within a confined space The predicted gas concentration value at the location, Indicating the strength of the gas leak source. Let be the gas diffusion coefficient. Discrete grid position Location of the leak source distance, For digital twin models at nodes Location Predicted gas concentration values at the location, number, Intensity of gas leak source The square of the L2 norm; For regularization system By solving the weighted cost function The minimum value is used to estimate the location of the leak source. and leakage intensity .
[0024] This step introduces turbulence intensity. This solves the problem of large calculation errors caused by large data fluctuations in traditional inversion algorithms in complex flow fields, and improves the robustness of source term localization.
[0025] S1.4 Leakage source location estimated based on S1.3 and leakage intensity Construct source term function : ; The global concentration field is then reconstructed using unsteady convection-diffusion equations: ; In the formula, For the first Gas concentration at each time step For the first Gas concentration at each time step For digital twin simulation time step, for Gas concentration distribution at time gradient vector, For gradient operators, For the first The spatial distribution gradient of gas concentration at each time step. To estimate the gas source term; by Using boundary constraints, we obtain all discrete meshes within the confined space. exist Gas concentration value at time By integrating all grid concentration values in grid order, a global gas concentration distribution matrix is obtained. : ; In the formula, For the first Discrete grid cells exist The gas concentration at that moment was high.
[0026] This step facilitates the transition from point data to field data. By calling... and This ensures the physical realism and dimensional accuracy of the numerical simulation, enabling the system to see the gas distribution within the sensor blind zone.
[0027] S1.5 Constructing a gradient prediction operator that includes the time derivative The concentration change rate was obtained by finite difference calculation: ; ; ; In the formula, The rate of change of concentration over time. For in position , The gas concentration at time , For in position , The gas concentration at time , For in position , The danger determination function at time t, For predicting time coefficients, The dangerous concentration threshold, Masking for high-risk areas, For the first Each grid in The danger threshold at any given moment.
[0028] This step facilitates the transition from point data to field data. This is achieved by calling the calculated... and This ensures the physical realism and dimensional accuracy of the numerical simulation, enabling the system to see the gas distribution within the sensor blind zone.
[0029] S2.1 Constructing a Discrete Mesh Set With sensor node set Static mapping relationship matrix between : ; In the formula, For the first Pre-defined deployment coordinates of each sensor node; During real-time operation, the high-risk area mask obtained based on S1.5 is used. Generate a preliminary candidate wake-up set : ; In the formula, This is the initial candidate wake-up set. For the first One sensor node, For the existence of masks belonging to high-risk areas The One grid; right Perform neighborhood expansion to obtain the expanded candidate wake-up set. : ; In the formula, To expand the candidate wake-up set, For the first One sensor node, For sensor nodes With sensor nodes Spatial distance between them This is the safety buffer distance threshold.
[0030] This step enables fast address translation from digital fields to physical devices. By introducing... Extending the neighborhood increases the system's robustness and ensures physical coverage of the edges of high-risk areas.
[0031] S2.2 For the expanded candidate wake-up set Any sensor node in Calculate the gas front arrival node Expected time Introducing local transport velocity : ; In the formula, For position Local transport velocity at the location, For position The reference wind speed vector at that location, This is the transport time scale coefficient. For position The effective diffusion coefficient at that location, The length of the grid feature; From the location of the leak source To sensor node Location Shortest fluid trace Integrate to obtain the estimated arrival time of the gas front. : ; In the formula, For the current moment, To the location of the leak source To sensor node Location The shortest fluid trace; according to Configure sensor nodes wake-up command time : ; In the formula, For sensor warm-up time, This represents network latency margin.
[0032] This step implements relay-style wake-up, utilizing... The prediction model was revised to ensure that the system can accurately predict the arrival time of toxic gas in complex flow fields (such as eddies and dead zones), achieving zero-delay capture.
[0033] S2.3 Constructing Sensor Nodes Activation utility function : ; In the formula, , , , For the utility function weights, For nodes Location exist Predicted gas concentration at any given time The time scale constant, For sensor nodes The remaining battery power, Fully charge the sensor nodes. This is the set of sensor nodes that are currently in operation. For sensor nodes With sensor nodes Spatial distance, It is a smoothing factor; Expanded candidate wake-up set Select a subset of wake-up nodes from the list to construct the final wake-up list. : ; In the formula, To expand the candidate wake-up set any subset, For subset The sum of the activation utility values of all sensor nodes in the system; And satisfy channel capacity constraints: ; In the formula, This is the preset channel capacity limit.
[0034] This step introduces... The system not only implemented on-demand wake-up but also time-priority allocation. This ensures that resources are preferentially allocated to nodes that are both dangerous and urgent, resolving channel congestion issues during multi-point outbreaks.
[0035] S2.4 gateway To the final wake-up list The node in the middle sends a wake-up frame: ; In the formula, For Bloom filter mask, This is the relative time offset; node The operating status is determined based on the signal strength and the Bloom filter verification result. ; In the formula, For sensor nodes The next work status, The preset signal strength threshold; In In this state, the node sets a wake-up timer based on the relative time offset and enters high-frequency sampling mode.
[0036] This step combines Bloom filters (compressed transmission) with relative time scheduling (no synchronization required) to achieve millisecond-level collaborative wake-up with extremely low overhead.
[0037] S3.1 Construct a weighted directed graph of the sensor network, for any two sensor nodes... , Define link transmission cost : ; In the formula, For risk aversion coefficient, For sensor nodes To sensor node Historical package delivery rate, For sensor nodes With sensor nodes Gas risk concentration in inter-linkage This is the critical risk threshold for gases. For router weight, This is the initial charge level of the sensor node. For sensor nodes Remaining battery power; Risk concentration on the link path pass obtained through linear sampling: ; In the formula, For sensor nodes With sensor nodes The straight path between them In order to be in At any given moment, the global gas concentration field In the middle, spatial location The corresponding predicted gas concentration; Construct a weighted adjacency matrix for the entire network based on signal strength thresholds. : ; In the formula, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between For sensor nodes To sensor node The received signal strength, This is the receiver sensitivity threshold; when When this occurs, it is considered that the physical link is unreachable, and communication is interrupted.
[0038] This step directly maps the spatial distribution of the physical field to the topological cost of the digital network. (Exponential term) The introduction of this technology causes the link cost in high-risk areas to increase non-linearly, forcing data flows to automatically bypass safe areas and construct a transmission path.
[0039] S3.2 Based on the sensor node set The set of dormant nodes is obtained. : ; from Selecting a set of relay nodes This minimizes the objective function: ; In the formula, The wake-up cost weighting coefficient, The fixed energy cost of waking up a relay node, This is the energy balance weighting coefficient. For environmental risk weighting coefficients, For relay nodes The remaining battery power, For relay nodes Location exist Predicted gas concentration at any given time This refers to the safe concentration threshold for the gas. And it satisfies connectivity constraints: ; active node set for: ; Depend on Constitutes the final active topology : .
[0040] This step addresses the transmission interruption problem caused by sparse sensing. It incorporates environmental risk constraints to ensure that the selected relay nodes not only maintain network connectivity but also remain viable long-term, significantly enhancing the resilience of the emergency network.
[0041] S3.3 Define Gateway Zero potential energy point, any sensor node The potential energy is determined by the cumulative link cost, and the calculation formula is: ; In the formula, For sensor nodes The potential energy value, For sensor nodes The set of all neighboring sensor nodes, For sensor nodes The potential energy value, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between; The next hop node is determined based on the potential energy of the data flow direction. The formula for calculating the set of next hop nodes is as follows: ; In the formula, For sensor nodes The set of next-hop nodes; Output the routing table for each source node. .
[0042] This step establishes a potential field gradient to naturally guide data flow to the gateway and forces the generation of multiple physically isolated paths, thus achieving highly reliable emergency communication.
[0043] S3.4 Define sensor nodes Dynamic latency requirements Dynamic latency requirements It is linked to the gas diffusion rate, and the calculation formula is: ; In the formula, Based on latency, The concentration change rate weighting coefficient, For sensor nodes Location The rate of change of concentration at that location; Define business priorities : ; In the formula, This is the priority weight coefficient. For nodes The actual concentration at the location The maximum gas concentration threshold; Based on the business priority of each node The sorting results are used to allocate dedicated guaranteed time slots to the highest priority nodes and ordinary transmission time slots to the remaining nodes. The time slot allocation results of all nodes are then compiled into a unified transmission time slot table. , For business priorities The highest-level node is allocated a dedicated guaranteed time slot, based on the receiver. Feedback adjustment node transmit power, the power adjustment formula is: ; In the formula, The transmission power at the previous moment. This is the power adjustment coefficient. For the target received signal strength, The received signal strength of the previous frame. For risk compensation power; Output the transmit power setting value for each sensor node. and transmission time slot table .
[0044] This step achieves deep coupling between the cross-layer protocol and the physical environment, directly translating the gas diffusion rate into latency requirements, ensuring that network transmission speed is always faster than the gas diffusion rate. Through closed-loop adjustments, signal fluctuations in explosion / leak environments are overcome, guaranteeing link stability.
[0045] S4.1 uses the remaining energy of each sensor node Delivery rate of each link Length of data queue for each sensor node ,definition The state vector of the entire network at time step : ; In the formula, For the first The remaining energy of each sensor node, For the first Delivery rate of each link, For the first The length of the data queue for each sensor node; Based on the sparsity of network state, a compressed sensing algorithm is used to reduce upload bandwidth, and a physical layer observation equation is established: ; In the formula, The state observation matrix, It is a sparse base. To observe noise; On the digital twin server side, through Norm minimization reconstructs the original state: ; ; In the formula, These are the measured values of the compressed state. The threshold for state reconstruction error; Output the reconstructed physical state of the entire network .
[0046] This step achieves comprehensive perception and low-power transmission of the physical network status. Through compressed sensing technology, it feeds back detailed health information of the physical network to the digital twin system with extremely low communication costs, providing a data foundation for subsequent closed-loop optimization.
[0047] S4.2 Define the model parameter vector to be corrected. : ; In the formula, For boundary wind speed, The turbulent diffusion coefficient; pass The model is evolved using a set of samples, and the evolution formula is as follows: ; In the formula, For the first Predicted parameters for a set of samples, These are the posterior analysis parameters from the previous time step. Evolutionary noise; Using the output of S1.2 Adjust parameters and introduce As an observation operator, the continuous field is mapped to the discrete sensor position, and the update formula is: ; In the formula, For the first Posterior analysis parameters of a set of samples, To assimilate the Kalman gain for the data, For solving convection-diffusion equations; Kalman gain calculate: ; In the formula, To predict the covariance matrix of the parameters, To observe the noise covariance matrix; right Posterior analysis parameters of a set of samples By taking the arithmetic mean and eliminating the influence of noise, the corrected optimal model parameters are obtained. .
[0048] This step establishes a physical consistency feedback loop between the real and virtual worlds. It addresses the issue of digital twin models gradually becoming distorted over long-term operation, ensuring that the model remains faithful to physical reality. This is a prerequisite for the entire system to make accurate predictions.
[0049] S4.3 with global concentration field The reconstructed physical state of the entire network System risk level Define the system state space as input. With action space : ; ; In the formula, For the global concentration field, This represents the average physical state of the entire network. , , For utility function weight adjustment, Adjustments made to mitigate risk. To avoid route adjustments, Adjustment for latency weights; Constructing a reward function : ; In the formula, , , As a reward weighting coefficient, For the sake of perception accuracy, For the overall network energy consumption, For alarm delay; The PPO algorithm is used to train the policy network. Output the optimal weight configuration: ; In the formula, This is the set of hyperparameters for the algorithm at the next time step. It is the set of basic hyperparameters.
[0050] This step enables intelligent strategic decision-making from virtual to real, dynamically adjusting algorithm weights based on energy consumption, risk, and latency to find the dynamic optimal solution between energy efficiency and safety.
[0051] S4.4 Defines System Operating Modes Based on the system state, adaptive mode switching is achieved, and the state transition equation is: ; In the formula, This is the highest concentration in the entire venue. This is the danger alarm threshold. This represents the link failure rate. The threshold for the link failure rate. The minimum remaining energy for all nodes in the network. The energy threshold for node death; against The mode is re-triggered in S3.2, and failed nodes are forcibly removed, resulting in an updated set of dormant nodes. : ; In the formula, For the updated set of dormant nodes, For a set of dormant nodes, This is the set difference operator. A subset of failed nodes; Output system current operating mode and system timing control parameters.
[0052] This step establishes the system's immune system and biological rhythms. Under normal conditions, the system operates at extremely low frequencies to maximize its lifespan; in emergency situations, it instantly switches to a high-frequency, highly active state to ensure zero latency. When a large-scale network failure is detected, it automatically enters recovery mode, using backup nodes to rebuild the network, thus endowing the system with the ability to survive in harsh environments.
[0053] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.
Claims
1. A digital twin driven gas monitoring network adaptive control method, characterized in that, Includes the following steps: S1. Construct a gridded mapping relationship between physical space and digital twin space, and use sparse sensor measured data to drive a virtual fluid dynamics model to obtain the global spatiotemporal state benchmark and benchmark flow field; S2. Based on the dynamic reconstruction results of the gas field, the high-risk area mask and the reference flow field parameters, the gas diffusion path and arrival time are predicted in advance by digital twin, and the dormant sensor nodes are accurately awakened. S3. Map gas concentration risk to communication link cost, and dynamically adjust network topology, relay nodes and link weights by combining wake-up node set to build a low-latency, high-redundancy adaptive physical network; S4. Establish a two-way feedback mechanism between the physical perception layer and the digital twin decision layer, feed back the physical network status to the digital twin model for parameter correction, and optimize the wake-up strategy and network parameters in reverse through the twin inference results to achieve system adaptive updates and fault self-healing.
2. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, The S1 includes S1.1 dividing a physical domain of a restricted space into a number of non-overlapping control volumes, all control volumes together forming a discrete mesh set , for any control volume defining a set of geometric properties and a flow field state vector : ； wherein is the geometric center coordinate of the th grid, is the physical volume of the th grid; ； wherein is the reference wind velocity vector in the first is the air density, is the turbulence intensity in the first is the effective diffusion coefficient; The reference flow field was obtained by solving the Navier-Stokes equations and coupled with... Turbulence model to calculate turbulence intensity and effective diffusion coefficient : ； ； ； ； In the formula, For Hamiltonian operators, air density, This is the reference wind speed vector within the grid. For turbulent kinetic energy, For turbulent viscosity, For turbulent kinetic energy, Prandtl number, The turbulent kinetic energy term is generated by the average velocity gradient. The turbulent kinetic energy dissipation rate, For the first Turbulent kinetic energy of each grid For the first Reference wind speed vector within each grid The model, The molecular diffusion coefficient is... For turbulent Schmidt number; Each grid , and By integrating the data, a baseline flow field database containing turbulence information is obtained. ; S1.2 settings At any given moment, all nodes in the physical mesh sensor network are traversed, and sensor nodes that are currently in working condition, collecting and uploading data normally are selected to form an active node set. For any node Define any node in State vector at time step : ； In the formula, For nodes exist The gas concentration collected at all times, For nodes exist The rate of change of gas concentration at time t; For nodes exist The original sampled signal at time t is subjected to improved Kalman filtering, and the filtering process follows the formula: ； ； ； In the formula, For nodes exist The predicted state value at time t, i.e., based on The state at time is derived The estimated state value at time step and the state vector Consistent in dimensions, including predicted gas concentrations and rates of change of concentration. Here is the state transition matrix. For nodes exist The state vector at time t, for Time-based process noise, For nodes exist The sampled value of the original gas concentration containing noise at any given time. For the observation matrix, The true state vector contains Real gas concentration at any given time and the actual concentration change rate , for Observation noise at any given moment For nodes exist The posterior estimate of the state at time t. Kalman gain for signal filtering; Output node is processed by filtering. High confidence concentration value after noise reduction ; High confidence concentration value after noise reduction According to the discrete grid set corresponding to the sensor node Integrating to form a sparse concentration vector : ； In the formula, To activate the node exist At time 1, the concentration estimate after Kalman filtering. This represents the total number of activated nodes.
3. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S1 includes S1.3, which is the sparse concentration vector output from S1.
2. Inverse solution of gas leak source strength and location Construct a weighted cost function : ； ； ； In the formula, For nodes The confidence weight coefficient, As the turbulence penalty factor, For nodes At discrete grid locations Turbulence intensity at that location This is a modified Gaussian plume model for confined spaces. For digital twin models at arbitrary discrete grid locations within a confined space The predicted gas concentration value at the location, Indicating the strength of the gas leak source. Let be the gas diffusion coefficient. Discrete grid position Location of the leak source distance, For digital twin models at nodes Location The predicted gas concentration value at the location, The regularization coefficient is . Intensity of gas leak source The square of the L2 norm; By solving the weighted cost function The minimum value is used to estimate the location of the leak source. and leakage intensity ; S1.4 Leak source location estimated based on S1.3 and leakage intensity Construct source term function : ； The global concentration field is then reconstructed using unsteady convection-diffusion equations: ； In the formula, For the first Gas concentration at each time step For the first Gas concentration at each time step For digital twin simulation time step, for Gas concentration distribution at time gradient vector, For gradient operators, For the first The spatial distribution gradient of gas concentration at each time step. To estimate the gas source term; by Using boundary constraints, we obtain all discrete meshes within the confined space. exist Gas concentration value at time By integrating all grid concentration values in grid order, a global gas concentration distribution matrix is obtained. : ； In the formula, For the first Discrete grid cells exist The gas concentration at a given time.
4. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S1 includes S1.5, which constructs a gradient prediction operator incorporating the time derivative. The concentration change rate was obtained by finite difference calculation: ； ； ； In the formula, The rate of change of concentration over time. For in position , The gas concentration at time , For in position , The gas concentration at time , For in position , The danger determination function at time t, For predicting time coefficients, The dangerous concentration threshold, Masking for high-risk areas, For the first Each grid in The danger threshold at any given moment.
5. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S2 includes S2.1 constructing a discrete mesh set. With sensor node set Static mapping relationship matrix between : ； In the formula, For the first Pre-defined deployment coordinates of each sensor node; During real-time operation, the high-risk area mask obtained based on S1.5 is used. Generate a preliminary candidate wake-up set : ； In the formula, This is the initial candidate wake-up set. For the first One sensor node, For the existence of masks belonging to high-risk areas The One grid; right Perform neighborhood expansion to obtain the expanded candidate wake-up set. : ； In the formula, To expand the candidate wake-up set, For the first One sensor node, For sensor nodes With sensor nodes Spatial distance between them This is the safety buffer distance threshold; S2.2 For the expanded candidate wake-up set Any sensor node in Calculate the gas front arrival node Expected time Introducing local transport velocity : ； In the formula, For position Local transport velocity at the location, For position The reference wind speed vector at that location, This is the transport time scale coefficient. For position The effective diffusion coefficient at that location, The length of the grid feature; From the location of the leak source To sensor node Location Shortest fluid trace Integrate to obtain the estimated arrival time of the gas front. : ； In the formula, For the current moment, To the location of the leak source To sensor node Location The shortest fluid trace; according to Configure sensor nodes wake-up command time : ； In the formula, For sensor warm-up time, This represents network latency margin.
6. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S2 includes S2.3 constructing sensor nodes. Activation utility function : ； In the formula, , , , For the utility function weights, For nodes Location exist Predicted gas concentration at any given time The time scale constant, For sensor nodes The remaining battery power, Fully charge the sensor nodes. This is the set of sensor nodes that are currently in operation. For sensor nodes With sensor nodes Spatial distance, It is a smoothing factor; Expanded candidate wake-up set Select a subset of wake-up nodes from the list to construct the final wake-up list. : ； In the formula, To expand the candidate wake-up set any subset, For subset The sum of the activation utility values of all sensor nodes in the system; And satisfy channel capacity constraints: ； In the formula, The preset channel capacity limit; S2.4 gateway To the final wake-up list The node in the middle sends a wake-up frame: ； In the formula, For Bloom filter mask, This is the relative time offset; node The operating status is determined based on the signal strength and the Bloom filter verification result. ； In the formula, For sensor nodes The next work status, The preset signal strength threshold; In In this state, the node sets a wake-up timer based on the relative time offset and enters high-frequency sampling mode.
7. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S3 includes S3.1 constructing a weighted directed graph of the sensor network, for any two sensor nodes... , Define link transmission cost : ； In the formula, For risk aversion coefficient, For sensor nodes To sensor node Historical package delivery rate, For sensor nodes With sensor nodes Gas risk concentration in inter-linkage This is the critical risk threshold for gases. For router weight, This is the initial charge level of the sensor node. For sensor nodes Remaining battery power; Risk concentration on the link path pass obtained by linear sampling: ； In the formula, For sensor nodes With sensor nodes The straight path between them In order to be in At any given moment, the global gas concentration field In the middle, spatial location The corresponding predicted gas concentration; Construct a weighted adjacency matrix for the entire network based on signal strength thresholds. : ； In the formula, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between For sensor nodes To sensor node The received signal strength, This is the receiver sensitivity threshold; when When this occurs, it is considered that the physical link is unreachable, and communication is interrupted; S3.2 Based on the sensor node set The set of dormant nodes is obtained. : ； from Selecting a set of relay nodes This minimizes the objective function: ； In the formula, The wake-up cost weighting coefficient, The fixed energy cost of waking up a relay node, This is the energy balance weighting coefficient. For environmental risk weighting coefficients, For relay nodes The remaining battery power, For relay nodes Location exist Predicted gas concentration at any given time This refers to the safe concentration threshold for the gas. active node set for: ； Depend on Constitutes the final active topology : ； S3.3 Define Gateway Zero potential energy point, any sensor node The potential energy is determined by the cumulative link cost, and the calculation formula is: ； In the formula, For sensor nodes The potential energy value, For sensor nodes The set of all neighboring sensor nodes, For sensor nodes The potential energy value, Weighted adjacency matrix Corresponding sensor nodes With sensor nodes Link weights between; The next hop node is determined based on the potential energy of the data flow direction. The formula for calculating the set of next hop nodes is as follows: ； In the formula, For sensor nodes The set of next-hop nodes; Output the routing table for each source node. .
8. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S3 includes S3.4 defining sensor nodes. Dynamic latency requirements Dynamic latency requirements It is related to the gas diffusion rate, and the calculation formula is: ； In the formula, Based on latency, The concentration change rate weighting coefficient, For sensor nodes Location The rate of change of concentration at that location; Define business priorities : ； In the formula, This is the priority weight coefficient. For nodes The actual concentration at the location The maximum gas concentration threshold; Based on the business priority of each node The sorting results are used to allocate dedicated guaranteed time slots to the highest priority nodes and ordinary transmission time slots to the remaining nodes. The time slot allocation results of all nodes are then compiled into a unified transmission time slot table. , For business priorities The highest-level node is allocated a dedicated guaranteed time slot, based on the receiver. Feedback adjustment node transmit power, the power adjustment formula is: ； In the formula, The transmission power at the previous moment. This is the power adjustment coefficient. For the target received signal strength, The received signal strength of the previous frame. For risk compensation power; Output the transmit power setting value for each sensor node. and transmission time slot table .
9. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S4 includes S4.1, which uses the remaining energy of each sensor node. Delivery rate of each link Length of data queue for each sensor node ,definition The state vector of the entire network at time step : ； In the formula, For the first The remaining energy of each sensor node, For the first Delivery rate of each link, For the first The length of the data queue for each sensor node; On the digital twin server side, through Norm minimization reconstructs the original state: ； Output the reconstructed physical state of the entire network ; S4.2 Define the model parameter vector to be corrected. : ； In the formula, For boundary wind speed, The turbulent diffusion coefficient; pass The model is evolved using a set of samples, and the evolution formula is as follows: ； In the formula, For the first Predicted parameters for a set of samples, These are the posterior analysis parameters from the previous time step. Evolutionary noise; Using the output of S1.2 Adjust parameters and introduce As an observation operator, the continuous field is mapped to the discrete sensor position, and the update formula is: ； In the formula, For the first Posterior analysis parameters of a set of samples, To assimilate the Kalman gain for the data, For solving convection-diffusion equations; right Posterior analysis parameters of a set of samples The arithmetic mean is taken to eliminate the influence of noise, and the corrected optimal model parameters are obtained. .
10. The adaptive control method for a gas monitoring network driven by a digital twin according to claim 1, characterized in that, S4 includes S4.3 with a global concentration field. The reconstructed physical state of the entire network System risk level Define the system state space as input. With action space : ； ； In the formula, For the global concentration field, This represents the average physical state of the entire network. , , For utility function weight adjustment, Adjustments made to mitigate risk. To avoid route adjustments, Adjustment for latency weights; The PPO algorithm is used to train the policy network. Output the optimal weight configuration: ； In the formula, This is the set of hyperparameters for the algorithm at the next time step. The set of basic hyperparameters; S4.4 Defines System Operating Modes Based on the system state, adaptive mode switching is achieved, and the state transition equation is: ； In the formula, This is the highest concentration in the entire venue. This is the danger alarm threshold. This represents the link failure rate. The threshold for the link failure rate. The minimum remaining energy for all nodes in the network. The energy threshold for node death; against The mode is re-triggered in S3.2, and failed nodes are forcibly removed, resulting in an updated set of dormant nodes. : ； In the formula, For the updated set of dormant nodes, For a set of dormant nodes, This is the set difference operator. A subset of failed nodes; Output system current operating mode and system timing control parameters.