A low-altitude airspace digital twin and multi-modal weighted fusion physical AI monitoring system and method

Abstract

This invention discloses a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system and method. It constructs a three-layer digital twin structure comprising a basic spatial layer, a dynamic layer, and a regional state layer, jointly inferring the target motion state and meteorological environmental variables within a global twin reference coordinate system. Spatiotemporal unified registration of heterogeneous observation data is achieved through sensing nodes. The system dynamically adjusts the observation noise covariance matrix and fusion weights of each sensor based on real-time meteorological variables, and uses Bayesian filtering incorporating meteorological constraints to complete the weighted update of the target's posterior state. Simultaneously, a dual-loop control architecture with decoupled edge fast loops and central slow loops is adopted, supporting autonomous operation of edge nodes during communication degradation and seamless stitching of the global state after recovery. Embedding meteorological mechanisms into the fusion process improves the absolute accuracy and system resilience of low-altitude monitoring under complex meteorological conditions, making it suitable for full-domain situational awareness and security assurance in high-density low-altitude scenarios such as urban air traffic.

Description

Technical Field

[0001] This invention relates to the fields of low-altitude airspace monitoring, digital twins, and multi-sensor data fusion technology. Specifically, it relates to a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system and method. Background Technology

[0002] With the development of the low-altitude economy and urban air traffic, the operational density of targets such as drones and aircraft in low-altitude airspace continues to increase, placing higher demands on the accuracy and reliability of airspace situational awareness. However, the low-altitude environment is complex and changeable, and the motion characteristics and observation features of targets such as micro-drones are easily affected by meteorological conditions such as wind field, rainfall, and visibility. Traditional monitoring methods are prone to problems such as tracking drift, false alarms, or missed alarms under complex weather conditions.

[0003] Current low-altitude monitoring systems mostly employ a distributed deployment of multiple sensors, including radar, photoelectric, and radio frequency sensors. Their data fusion is largely limited to the result level, such as track stitching, and fails to achieve joint estimation of the underlying state under a unified spatiotemporal reference. Specifically, existing technologies suffer from the following main shortcomings:

[0004] First, regarding target state estimation, most existing systems only use meteorological data as external auxiliary information and do not explicitly incorporate it into the target's dynamic evolution model. This results in the system's trajectory prediction and state estimation often deviating from the target's actual physical motion process under conditions such as strong winds and heavy rain.

[0005] Secondly, in terms of multi-sensor data fusion, existing systems often use fixed values ​​for multi-modal observation fusion weights, or simply smooth them based on historical observation quality, failing to establish a dynamic correlation with real-time meteorological conditions. Therefore, when encountering sudden severe environments such as rainfall attenuation or low visibility, the system struggles to adaptively adjust the confidence levels of the interfered sensors, easily leading to divergence in the fusion results.

[0006] Finally, regarding system architecture and operation and maintenance, existing low-altitude monitoring platforms heavily rely on central nodes for centralized computation, with edge nodes only handling data acquisition and simple preprocessing. In complex scenarios such as limited communication bandwidth or link interruptions, edge nodes struggle to maintain stable closed-loop tracking independently, easily leading to localized blind spots. Furthermore, existing systems lack a closed-loop iterative mechanism based on digital twins and offline simulation, making it difficult to continuously optimize model parameters for special operating conditions such as extreme weather and complex trajectories, thus failing to meet the safety requirements of high-density low-altitude operations. Summary of the Invention

[0007] In view of the above problems, the purpose of this invention is to provide a digital twin and multimodal weighted fusion monitoring system and method for low-altitude airspace, which embeds meteorological and physical constraints into the state estimation and data fusion process to improve the accuracy and robustness of intelligent monitoring of low-altitude airspace under complex meteorological conditions.

[0008] According to one aspect of the present invention, a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system is provided, comprising: a low-altitude airspace digital twin subsystem, used to initialize a global twin reference coordinate system and construct a three-layer digital twin structure including a basic spatial layer, a dynamic layer, and a regional state layer, so as to jointly express and collaboratively infer the joint state vector composed of the target entity motion state and meteorological environmental variables in a unified state space; a sensing data subsystem, including at least one standardized sensing node, used to acquire multi-source heterogeneous observation data and uniformly convert the observation data and the corresponding measurement error model to the global twin reference coordinate system; and a physical intelligence core subsystem, which is connected to the low-altitude airspace digital twin subsystem and the sensing data subsystem respectively. The system is interconnected and used to extract real-time meteorological variables from the dynamic layer. Based on a preset meteorological-noise mapping relationship, it dynamically adjusts the observation noise covariance matrix of each modal sensor and adaptively corrects the fusion weights of multimodal observation data accordingly. It uses recursive Bayesian filtering to achieve weighted updates of the target state. The physical intelligence core subsystem is equipped with a dual-loop control module for decoupling edge and center communication, including an edge fast loop running at a first preset frequency cycle and a center slow loop running at a second preset frequency cycle. The standardized node collaboration and offline simulation optimization subsystem is used to monitor the communication status in real time, trigger the standardized sensing nodes to enter the edge autonomous mode under communication degradation conditions, and perform consistent smooth stitching of local and global tracks after communication is restored.

[0009] Optionally, in the low-altitude airspace digital twin subsystem: the basic spatial layer is used to store digital maps, regional electronic boundaries, and sensor three-dimensional spatial layout information; the dynamic layer is used to incorporate the target motion state vector containing position and velocity and the meteorological environmental variable vector containing wind field, rainfall, and visibility into the same dynamic evolution equation for collaborative deduction; the regional state layer is used to dynamically divide the global space into measurement area, prediction area, and unknown area based on the posterior estimated covariance matrix trace and observation coverage timeliness.

[0010] Optionally, the standardized sensing node includes: a low-altitude surveillance radar, a radio frequency detection module, an optical gimbal, and a meteorological sensing module; and an edge computing unit, which is physically deployed near the low-altitude surveillance radar and directly connected to it, serving as the computing power host of the edge loop.

[0011] Optionally, in the physical intelligence core subsystem: the edge fast loop is used to complete the target posterior state update based on local sensor data and directly generate the servo drive closed-loop tracking command of the optical gimbal; the central slow loop is used to receive the state data uploaded by each edge node, perform regional multi-node cross-track map optimization alignment and global collision avoidance risk assessment.

[0012] Optionally, the dynamic adjustment of the observation noise covariance matrix of each modal sensor based on the preset meteorological-noise mapping relationship includes: when the monitored rainfall rate increases to a first preset threshold, the diagonal elements of the observation noise covariance matrix are amplified by a preset ratio; when the monitored visibility decreases to a second preset threshold, the optical angle error covariance matrix is ​​amplified by a preset attenuation function; by amplifying the noise covariance of the interfered sensor, its Kalman gain weight in the recursive Bayesian filter is adaptively reduced.

[0013] According to another aspect of the present invention, a method for monitoring low-altitude airspace digital twins and multimodal weighted fusion physical AI is provided, comprising the following steps:

[0014] S1, Construct a joint twin space: Initialize the global twin reference coordinate system through the low-altitude airspace digital twin subsystem, construct a three-layer digital twin structure, and jointly infer the motion state of the target entity and meteorological environmental variables;

[0015] S2, Multimodal sensing and spatiotemporal normalization: Collect multimodal observation data and local meteorological environmental variables, and register the heterogeneous data and spatiotemporal error model to the global twin reference coordinate system;

[0016] S3, Meteorological Adaptive Weight Adjustment and Weighted Fusion: Extract real-time meteorological variables, dynamically calculate the observation noise covariance matrix of each modality and correct the fusion weights, and use a recursive Bayesian filter to complete the weighted update of the target posterior state.

[0017] S4, Dual-loop Cooperative Control and Situation Feedback Mapping: Executes dual-loop control logic, with the edge fast loop outputting sensor closed-loop tracking commands and the central slow loop completing the macroscopic association of multiple nodes; at the same time, the grid attribute division of the regional state layer is updated in real time based on the updated posterior estimate covariance.

[0018] S5, Autonomous Operation and Model Evolution: Monitors communication status and performs edge autonomous operation and smooth stitching of tracks after communication recovery; at the same time, it synchronizes the scene data captured online to the offline simulation environment for parameter optimization and regularly updates the model parameters of the physical intelligence core subsystem.

[0019] Optionally, in S1: the joint state vector is represented as ,in Let be the target motion state vector. The vector represents meteorological environmental variables; the physical coupling state transition equation executed in the dynamic layer is: ,in This is the wind field mapping matrix, used to achieve physical wind deflection compensation for target prediction.

[0020] Optionally, in S4, the spatial grid attribute division of the regional state layer includes: the measurement region updated by recent high-confidence observations; the prediction region that is beyond the sensor detection range or physically blocked and whose prediction is maintained solely by the dynamic evolution equation; and the unknown region where the signal-to-noise ratio is below a preset threshold or the uncertainty covariance exceeds a preset upper limit.

[0021] Optionally, in S5: after communication is restored, an asynchronous state synchronization mechanism is used to perform consistency correction between the local independent track and the global macro track during the disconnection period, so as to achieve a smooth transition of situation data.

[0022] Optionally, in S5: the evolved meteorological-noise mapping model parameters and filter parameters generated after parameter iterative optimization are periodically updated to the edge computing unit and central server of the physical intelligence core subsystem through a network distribution mechanism.

[0023] Compared with existing technologies, the low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system and method provided by this invention have the following beneficial effects: (1) This invention explicitly embeds meteorological environmental variables such as wind field and rainfall into the joint evolution equation of the dynamic layer of the digital twin. By constructing a state estimation architecture that integrates meteorological constraints and data-driven approaches, the accuracy and reliability of target trajectory prediction and state estimation are improved. (2) This invention achieves smooth transfer of cross-modal confidence by dynamically adjusting the observation noise covariance matrix of each modal sensor, thereby improving the tracking stability of low-altitude targets under complex weather conditions. (3) This invention adopts a dual-loop control mechanism with decoupled edge fast loop and central slow loop. The front end relies on edge computing power to realize closed-loop servo tracking and aiming, while the back end performs cross-node cross-track map optimization, which simultaneously satisfies the constraints of low-latency response of target tracking and global situation assessment at the system level. (4) This system supports switching to edge autonomous mode when communication is downgraded, and performs consistency correction between local independent tracks and central macro tracks after communication is restored, avoiding state jumps when reconnecting to the network and enhancing the resilience of system operation. (5) This invention synchronizes the scene data captured online to the offline virtual space for algorithm testing and parameter optimization, and regularly distributes updates through the network mechanism, providing a technical implementation path for model iteration and dynamic parameter updates of the monitoring system. Attached Figure Description

[0024] Figure 1 A schematic diagram of the overall architecture of a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system provided in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the standardized sensing node hardware physical topology provided in an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of the three-layer logical architecture of the low-altitude airspace digital twin subsystem provided in this embodiment of the invention;

[0027] Figure 4 A schematic diagram of the dual-loop control structure and data flow of the physical intelligence core subsystem provided in this embodiment of the invention;

[0028] Figure 5 A flowchart of a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring method provided in an embodiment of the present invention.

[0029] Explanation of reference numerals in the attached figures:

[0030] 10-Low-altitude airspace digital twin subsystem; 11-Basic space layer; 12-Dynamic layer; 13-Regional state layer; 20-Sensing data subsystem; 21-Standardized sensing node; 211-Low-altitude surveillance radar (or multi-functional radar); 212-Radio frequency detection module; 213-Optical gimbal; 214-Meteorological sensing module; 215-Edge computing unit; 22-Spatiotemporal registration and normalization module; 30-Physical intelligence core subsystem; 31-Edge fast loop; 32-Center slow loop; 40-Standardized node collaboration and offline simulation optimization subsystem. Detailed Implementation

[0031] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0032] It should be noted that most of the accompanying drawings of this invention are logical structure diagrams, system architecture diagrams, or method flowcharts. These are primarily used to clearly illustrate the module division, data flow, and control logic of the invention, and do not represent the actual size, absolute position, or proportional relationship of the physical hardware. In the various drawings, the same or similar modules are represented by similar reference numerals. Furthermore, to highlight the core innovations of this invention, certain well-known low-level communication interfaces or conventional computing components may not be shown in the drawings.

[0033] Many specific details of the invention, such as specific subsystem partitioning, algorithm derivation formulas, and dual-loop control logic, are described below to provide a clearer understanding of the invention. However, those skilled in the art will understand that the invention can be practiced without relying on these specific details. For example, those skilled in the art can make equivalent substitutions for parameters of certain specific algorithms, specific models of hardware nodes, or communication protocols without departing from the inventive concept.

[0034] It should be noted first that the "Physical AI" mentioned in this embodiment specifically refers to a hardware and software collaborative computing architecture that deeply integrates prior physical knowledge such as target dynamics models and meteorological environment evolution mechanisms with a data-driven recursive Bayesian filtering and multimodal weighted fusion framework. In the specific system architecture, the core processing logic and model deduction of Physical AI are specifically carried out and executed by the physical intelligence core subsystem described later.

[0035] Figure 1 A schematic diagram of the overall architecture of a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to an embodiment of the present invention is shown. Figure 1 As shown, the overall architecture of the system mainly comprises four core components: a low-altitude airspace digital twin subsystem 10, a sensing data subsystem 20, a physical intelligence core subsystem 30, and a standardized node collaboration and offline simulation optimization subsystem 40. Logically, the system constructs a unified data and spatial foundation through the low-altitude airspace digital twin subsystem 10; in the control chain, the physical intelligence core subsystem 30 achieves edge-center collaboration through a dual-ring decoupling architecture. Data interaction and command issuance between the subsystems are achieved through wired or wireless communication networks.

[0036] Figure 2 A schematic diagram of the hardware physical topology of the standardized sensing node 21 in the sensing data subsystem 20 according to an embodiment of the present invention is shown. Figure 2 As shown, to ensure efficient execution of underlying computing power and low-latency data transmission, the standardized sensing node 21 is configured as a highly integrated modular physical detection base station, and adopts a hardware direct-connection architecture with the low-altitude surveillance radar 211 as the core. Specifically, the front end of the node integrates a low-altitude surveillance radar 211 (or a multi-functional radar), a radio frequency detection module 212, an optical gimbal 213, and a meteorological sensing module 214.

[0037] Specifically, the standardized sensing node 21 also includes an edge computing unit 215. This edge computing unit 215, serving as the computing power host for the "edge loop" in the physical intelligence core subsystem 30, is physically deployed near the low-altitude surveillance radar 211. For example, the edge computing unit 215 can be conformally integrated with the low-altitude surveillance radar 211 or directly connected via a high-speed backplane bus. High-frequency, high-bandwidth raw point or track data from the low-altitude surveillance radar 211 is directly injected into the edge computing unit 215, while auxiliary modal data such as the optical gimbal 213 and the radio frequency detection module 212 are aggregated to the computing unit 215 via the node's local area network (LAN).

[0038] Regarding the mechanism by which the meteorological sensing module 214 acquires meteorological physical parameters, this embodiment provides two compatible hardware configuration modes: Mode 1: When the standardized sensing node 21 adopts a multi-functional radar (such as a multi-functional phased array radar), the radar simultaneously analyzes the environmental body scattered echoes to extract meteorological information such as rainfall and wind fields while performing low-altitude target monitoring. This raw meteorological information is directly output internally and provided to the meteorological sensing module 214. Mode 2: When the standardized sensing node 21 adopts a traditional monitoring radar (with only target detection function), the meteorological sensing module 214 directly acquires local data through its integrated micro-meteorological station hardware (such as anemometers, rain gauges, visibility meters, etc.) and / or reads external regional meteorological network data and third-party meteorological sensor node information in real time through the external communication interface of the edge computing unit 215 to supplement the input of environmental meteorological status.

[0039] Furthermore, the multi-source heterogeneous data collected by each front-end sensor are uniformly fed into the spatiotemporal registration and normalization module 22 deployed within the edge computing unit 215. This spatiotemporal registration and normalization module utilizes the intrinsic and extrinsic parameter calibration matrices of each sensor and network time synchronization mechanisms (such as PTP Precision Time Protocol or NTP Network Time Protocol) to uniformly transform and align the measurement error model and observation data to a global twin reference coordinate system, thereby providing a standardized joint data source for the subsequent weighted fusion of physical AI.

[0040] Figure 3 A schematic diagram of the three-layer logical architecture of a low-altitude airspace digital twin subsystem 10 according to an embodiment of the present invention is shown. Figure 3 As shown, the low-altitude airspace digital twin subsystem 10 constructs a three-layer logical architecture comprising a basic spatial layer 11, a dynamic layer 12, and a regional state layer 13.

[0041] The basic spatial layer 11 stores digital maps, regional electronic boundaries, and three-dimensional spatial layout information of multimodal sensors. In the dynamic layer 12, to achieve the co-evolution of the target and the meteorological environment, this embodiment constructs a joint state vector with deep meteorological coupling. Specifically, it is expressed as:

[0042]

[0043] in, This represents a discrete-time index. Let the target's three-dimensional motion state vector be defined as:

[0044]

[0045] In the formula, Let the target's three-dimensional position coordinates be in the global twin reference coordinate system. These are the corresponding three-dimensional velocity components. The meteorological environmental variable vector obtained by the meteorological sensing module 214 or an external interface is defined as:

[0046]

[0047] In the formula, For rainfall rate, Visibility distance, These are the three-dimensional wind field vector components.

[0048] Based on the aforementioned joint state vector, the target dynamics evolution equation (i.e., state transition equation) executed by the physical intelligence core subsystem 30 in the dynamic layer 12 is as follows:

[0049]

[0050] In the formula, This is the state transition matrix; This is the wind field mapping matrix, used to directly map the wind field vector and superimpose it onto the target kinematics prediction to achieve underlying physical wind deflection compensation for the target prediction. Let be the process noise covariance matrix. This equation establishes the basis for incorporating meteorological and physical priors into the derivation of this system.

[0051] Figure 4 A schematic diagram illustrating the dual-loop control structure and data flow of the physical intelligence core subsystem 30 according to an embodiment of the present invention is shown. Figure 4 As shown, the physical intelligence core subsystem 30 performs adaptive weighted fusion based on the meteorological-noise mapping relationship. This physical intelligence core subsystem 30 designs it as a dynamic function of meteorological variables. :

[0052]

[0053] In the formula, This indicates the construction of a block diagonal matrix. This is the measurement noise covariance matrix of the low-altitude surveillance radar 211. When encountering heavy rain (i.e., the monitored rainfall rate)... Greater than the first preset threshold When this occurs, the system adaptively enlarges its diagonal elements according to a preset ratio. As a specific embodiment, this enlargement logic can be implemented using the following linear penalty function:

[0054]

[0055] in, For the nominal rated noise, This is the preset proportional adjustment coefficient.

[0056] Similarly, This is the noise matrix of the optical gimbal 213. When visibility... Drop to the second preset threshold In the following case, the system amplifies its angular error matrix according to a preset attenuation function. As a specific embodiment, this amplification logic can be implemented using the following exponential attenuation function:

[0057]

[0058] in, The nominal optical noise, This is the preset attenuation control coefficient. Furthermore, The noise matrix of the radio frequency detection module 212 is less affected by weather conditions and is set as a relatively stable constant in this embodiment.

[0059] Subsequently, the edge loop 31 deployed at the front end performs multimodal adaptive weighted fusion to obtain the fused target posterior state estimate. :

[0060]

[0061] In the formula, This represents the total number of sensor modes. For the first Local state estimation for each mode. Weights Strongly correlated with the calculated current noise matrix: the greater the interference from severe weather, the amplified the noise variance of the sensor, and the lower the assigned Kalman gain weight; the system then smoothly transfers the confidence level to the undisturbed auxiliary mode, with all weights satisfying .

[0062] At the control output end, the edge fast loop 31 deployed in the edge computing unit 215 directly outputs servo tracking commands to the optical gimbal 213 based on the aforementioned high-frequency fused data. Simultaneously, the central slow loop 32 deployed in the backend server asynchronously receives edge node data to complete regional-level cross-track map optimization alignment and collision avoidance risk assessment. Furthermore, the posterior estimated covariance matrix trace output after fusion is fed back to the regional state layer 13 in real time to dynamically delineate measurement areas driven by recent high-confidence observations, prediction areas maintained solely by evolutionary equations, and unknown areas with uncertainties exceeding a preset upper limit.

[0063] The standardized node collaboration and offline simulation optimization subsystem 40 is responsible for state collaboration and operation and maintenance during system operation. Specifically, this subsystem 40 monitors the communication link status between the central server and the edge computing units in real time. When a communication link degradation or interruption is detected, the edge fast loop 31 is triggered to enter autonomous mode, independently maintaining closed-loop tracking of the target based on local computing power. After the communication link is restored, the consistency and smoothing of the edge local independent track and the central global macro track are performed.

[0064] Simultaneously, the standardized node collaboration and offline simulation optimization subsystem 40 synchronizes online operational data containing extreme weather (such as strong winds and heavy rain) scenario characteristics to the offline digital twin platform for stress testing of the multimodal weighted fusion algorithm. Furthermore, through the OTA (Over-the-Air) network mechanism, the optimized and iteratively optimized meteorological-noise model parameters (such as the proportional adjustment coefficient α and attenuation control coefficient β in the aforementioned embodiment) are periodically distributed and updated to the physical intelligence core subsystem 30, thereby achieving continuous evolution of the system model parameters.

[0065] Figure 5 A flowchart illustrating a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring method according to an embodiment of the present invention is shown. Figure 5 As shown, this method corresponds to the system architecture described above and mainly includes the following steps:

[0066] S1. Constructing a Joint Twin Space: A global twin reference coordinate system is initialized through the low-altitude airspace digital twin subsystem 10, constructing a three-layer digital twin structure comprising a basic spatial layer, a dynamic layer, and a regional state layer. Within this unified state space, the motion state of the target entity and meteorological environmental variables are jointly expressed and extrapolated.

[0067] S2, Multimodal Sensing and Spatiotemporal Normalization: The sensing data subsystem 20 continuously collects multi-source observation and local meteorological data. Subsequently, the spatiotemporal registration and normalization module 22 completes the registration and unification of heterogeneous data and spatiotemporal error models in the global twin reference coordinate system.

[0068] S3, Meteorological Adaptive Weight Adjustment and Weighted Fusion: The physical intelligence core subsystem 30 dynamically calculates the observation noise covariance matrix for each modality based on extracted real-time meteorological variables. On this basis, it dynamically corrects the fusion weights according to the aforementioned formula and uses a recursive Bayesian filter to complete the weighted update of the target posterior state.

[0069] S4, Dual-loop Cooperative Control and Situation Feedback Mapping: Executes the dual-loop control logic of edge fast loop 31 and center slow loop 32; at the same time, based on the updated posterior covariance matrix trace, updates the grid attributes of the regional state layer 13 in real time (such as the division of the measurement area, prediction area and unknown area), and outputs the corresponding risk and credibility assessment.

[0070] S5, Autonomous Operation and Model Evolution: During system operation, the network status is monitored through standardized node collaboration and offline simulation optimization subsystem 40. When network degradation is detected, edge node autonomy is triggered; and after communication is restored, a consistent smooth stitching of the global state is performed. Simultaneously, model testing based on offline simulation is executed, and the iteratively updated parameters are fed back to the online system.

[0071] As described above, these embodiments of the present invention do not exhaustively cover all details, nor do they limit the invention to the specific embodiments described. Clearly, many modifications and variations can be made based on the above description. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to effectively utilize the invention and its modifications. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system, characterized in that, include: The low-altitude airspace digital twin subsystem (10) is used to initialize the global twin reference coordinate system and construct a three-layer digital twin structure including a basic space layer (11), a dynamic layer (12) and a regional state layer (13) to jointly express and collaboratively deduce the joint state vector composed of the target entity motion state and meteorological environmental variables in a unified state space. The sensing data subsystem (20) includes at least one standardized sensing node (21) for acquiring multi-source heterogeneous observation data and uniformly converting the observation data and the corresponding measurement error model to the global twin reference coordinate system. The physical intelligence core subsystem (30) is connected to the low-altitude airspace digital twin subsystem (10) and the sensing data subsystem (20) respectively. It is used to extract real-time meteorological variables from the dynamic layer (12), dynamically adjust the observation noise covariance matrix of each modal sensor based on the preset meteorological-noise mapping relationship, and adaptively correct the fusion weight of multimodal observation data accordingly. It uses recursive Bayesian filtering to realize the weighted update of the target state. The physical intelligence core subsystem (30) is equipped with a dual-loop control module with decoupled edge and center communication, including an edge fast loop (31) running at a first preset frequency period and a center slow loop (32) running at a second preset frequency period, wherein the first preset frequency is greater than the second preset frequency. The standardized node collaboration and offline simulation optimization subsystem (40) is used to monitor the communication status in real time, trigger the standardized sensing node (21) to enter the edge autonomous mode under the condition of communication degradation, and perform consistent smooth stitching of local track and global track after communication is restored.

2. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 1, characterized in that, In the low-altitude airspace digital twin subsystem (10): The basic spatial layer (11) is used to store digital maps, regional electronic boundaries, and sensor three-dimensional spatial layout information; The dynamic layer (12) is used to incorporate the target motion state vector containing position and velocity, and the meteorological environmental variable vector containing wind field, rainfall and visibility into the same dynamic evolution equation for collaborative deduction. The regional state layer (13) is used to dynamically divide the global space into measurement region, prediction region and unknown region based on the posterior estimated covariance matrix trace and observation coverage time.

3. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 1, characterized in that, The standardized sensing node (21) includes: Low-altitude surveillance radar (211), radio frequency detection module (212), optical gimbal (213) and weather sensing module (214). The edge computing unit (215) is physically deployed near the low-altitude surveillance radar (211) and directly connected to it, serving as the computing power host of the edge loop (31).

4. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 1, characterized in that, In the physical intelligence core subsystem (30): The edge loop (31) is used to update the target posterior state based on local sensor data and directly generate servo drive closed-loop tracking commands for the optical gimbal (213). The central slow loop (32) is used to receive the status data uploaded by each edge node, and to perform regional multi-node cross track map optimization alignment and global collision avoidance risk assessment.

5. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 1, characterized in that, The dynamic adjustment of the observation noise covariance matrix of each modal sensor based on the preset meteorological-noise mapping relationship includes: When the monitored rainfall rate increases to the first preset threshold, the diagonal elements of the observation noise covariance matrix are amplified according to a preset proportional adjustment relationship.

6. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 1, characterized in that, The dynamic adjustment of the observation noise covariance matrix of each modal sensor based on the preset meteorological-noise mapping relationship also includes: When the monitored visibility drops to a second preset threshold, the angular error covariance matrix of the optical gimbal is amplified according to a preset attenuation function relationship.

7. A low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system according to claim 5 or 6, characterized in that, The adaptively corrected fusion weights for the multimodal observation data include: By amplifying the noise covariance of the weather-affected sensor, the Kalman gain weight of the weather-affected sensor in the recursive Bayesian filter is adaptively reduced.

8. A method for monitoring low-altitude airspace using digital twins and multimodal weighted fusion physical AI, characterized in that, The system, applied to a low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring system as described in any one of claims 1-7, includes the following steps: S1, Construct a joint twin space: Initialize the global twin reference coordinate system through the low-altitude airspace digital twin subsystem, construct a three-layer digital twin structure, and jointly infer the motion state of the target entity and meteorological environmental variables; S2, Multimodal sensing and spatiotemporal normalization: Collect multimodal observation data and local meteorological environmental variables, and register the heterogeneous data and spatiotemporal error model to the global twin reference coordinate system; S3, Meteorological Adaptive Weight Adjustment and Weighted Fusion: Extract real-time meteorological variables, dynamically calculate the observation noise covariance matrix of each modality and correct the fusion weights, and use a recursive Bayesian filter to complete the weighted update of the target posterior state. S4, Dual-loop Cooperative Control and Situation Feedback Mapping: Executes dual-loop control logic, with the edge fast loop outputting sensor closed-loop tracking commands and the central slow loop completing the macroscopic association of multiple nodes; at the same time, the grid attribute division of the regional state layer is updated in real time based on the updated posterior estimate covariance. S5, Autonomous Operation and Model Evolution: Monitor network communication status and execute autonomous operation of edge nodes when communication degrades, and perform consistent smooth stitching of local and global tracks after communication is restored; at the same time, synchronize the scene data captured online to the offline simulation environment for parameter iterative optimization, and update the model parameters of the physical intelligence core subsystem.

9. The method for monitoring low-altitude airspace using digital twins and multimodal weighted fusion physical AI according to claim 8, characterized in that, In S1: The joint state vector is represented as ,in Let be the target motion state vector. A vector of meteorological and environmental variables; The physical coupling state transition equations executed in the dynamic layer are as follows: ,in Here is the state transition matrix. This is the wind field mapping matrix, used to achieve physical wind deflection compensation for target prediction. Let be the process noise covariance matrix.

10. A low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring method according to claim 8, characterized in that, In S4, the spatial grid attribute division of the region state layer includes: The measurement area is updated based on recent high-confidence observations; The prediction area is beyond the sensor's detection range or is physically blocked, and the prediction is maintained solely by the dynamic evolution equation; And unknown regions where the signal-to-noise ratio is below a preset threshold or the uncertainty covariance exceeds a preset upper limit.

11. The method for monitoring low-altitude airspace using digital twins and multimodal weighted fusion physical AI according to claim 8, characterized in that, In S5: After communication is restored, an asynchronous state synchronization mechanism is used to perform consistency correction between local independent tracks and global macro tracks during the disconnection period, so as to achieve a smooth transition of situation data.

12. The low-altitude airspace digital twin and multimodal weighted fusion physical AI monitoring method according to claim 8, characterized in that, In S5: the evolved meteorological-noise mapping model parameters and filter parameters generated after parameter iterative optimization are periodically synchronized to the edge computing unit and central server of the physical intelligence core subsystem through a network distribution mechanism.

Images