A low-altitude intelligent networking environment situation awareness method based on machine learning

Abstract

The application relates to the technical field of low-altitude intelligent networking environment sensing, in particular to a low-altitude intelligent networking environment situation awareness method based on machine learning; the method comprises the following steps: constructing a multi-source heterogeneous sensing system to collect flight, communication, meteorological and spatial multi-dimensional data, generating an environment state element sequence containing spatial attributes and evolution trend; establishing a machine learning evolution model to learn the environment coupling law and predict the future state; generating a multi-path situation evolution track based on the state transition probability, and dynamically correcting in combination with the spatial correlation and credibility evaluation; constructing an active sensing optimization mechanism, adjusting the collection cycle and edge computing resources according to the prediction demand, and correcting the prediction and feeding back the model by using real-time feedback. The application realizes accurate and continuous deduction of the low-altitude environment situation and closed-loop self-adaptive optimization of the sensing strategy, and enhances the accuracy of situation awareness and the resource scheduling efficiency in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of low-altitude intelligent network environment perception technology, specifically to a low-altitude intelligent network environment situational perception method based on machine learning. Background Technology

[0002] With the rapid development of the Low-Altitude Intelligent Network and the increasing activity of low-altitude economic activities, the number of aircraft in the low-altitude area has increased dramatically, flight path structures have become increasingly complex, and the pressure on communication links has continued to rise. Meanwhile, the deep interplay of factors such as sudden changes in meteorological conditions and spatial constraints imposed by urban built-up environments has resulted in a low-altitude operating environment exhibiting complex characteristics such as strong dynamism, multi-dimensional coupling, and high uncertainty. Environmental situational awareness, as a core foundational capability for ensuring low-altitude flight safety, optimizing flight path planning, supporting communication resource scheduling, and risk early warning, directly impacts the overall operational efficiency of the Low-Altitude Intelligent Network in terms of accuracy, continuity, and adaptability. Currently, the technical means for low-altitude environmental monitoring and state perception are constantly being enriched, and various types of sensing equipment have been widely applied in areas such as flight status acquisition, communication quality monitoring, micro-meteorological observation, and spatial obstacle identification, providing important data sources and preliminary analytical capabilities for low-altitude environmental cognition. Simultaneously, the exploration of machine learning methods in time series analysis and state prediction has provided valuable technical reserves for uncovering the laws governing environmental evolution.

[0003] Chinese invention patent CN118280168B discloses a method and system for low-altitude airspace management based on integrated sensing. The method includes the following steps: constructing a semantic feature map of the target low-altitude airspace; performing spatiotemporal reasoning analysis of the low-altitude environment based on the semantic feature map to obtain low-altitude environment representation data; performing real-time semantic segmentation on the low-altitude environment representation data to obtain a real-time semantic understanding model; performing online incremental learning based on the real-time semantic understanding model to obtain an adaptive semantic understanding model; and performing low-altitude geometric target detection based on the adaptive semantic understanding model to obtain a dynamic semantic map of the low-altitude environment.

[0004] However, changes in the low-altitude environment are not simply the sum of various factors, but a complex evolutionary process in which flight behavior, communication fluctuations, weather disturbances, and space occupation interact and constrain each other. Static analysis in a single dimension is no longer sufficient to meet the need for accurate understanding of future trends. There is an urgent need to establish a comprehensive method at the system level that can integrate heterogeneous data, capture evolutionary patterns, and achieve dynamic adaptation of perception strategies. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a low-altitude intelligent network environment situational awareness method based on machine learning.

[0006] The technical solution of this invention: a low-altitude intelligent network environment situational awareness method based on machine learning, comprising the following specific implementation steps: S1. Construct a low-altitude intelligent network multi-source heterogeneous sensing system to synchronously collect multi-dimensional data and uniformly process differences in data sources, sampling periods and spatial distribution. Combine time window division, spatial influence weight calculation and environmental change feature extraction to convert continuously changing low-altitude environmental information into a sequence of environmental state primitives with spatial attributes, state attributes and evolutionary trends. S2. Construct a machine learning analysis model for learning the laws of environmental evolution. By dynamically analyzing the current state, direction of change, operational behavior and related environmental factors in the environmental state primitives, a state representation that can describe the process of low-altitude environmental change is formed. The machine learning model is used to learn the coupling relationship between flight activities, communication status, meteorological disturbances and spatial changes, establish the environmental state evolution mapping relationship, and combine trend calculation and feedback correction to complete the prediction of the future direction of low-altitude environmental change and model adaptive optimization. S3. Construct the low-altitude environment situation evolution trajectory. By extracting the relationship between environmental state changes within a continuous time window, construct state transition units and calculate the probability of environmental state transition using historical state transition patterns. Generate future multi-path situation evolution trajectories. Combine the spatial correlation between different low-altitude areas to analyze the propagation impact of environmental changes. Optimize the prediction results through trajectory credibility evaluation and feedback correction to complete the dynamic extrapolation of the future change process of the low-altitude environment. S4. Construct a future situation-driven proactive perception optimization mechanism. Calculate regional proactive perception demand indicators by analyzing the future environmental state change trends of different regions. Based on the demand results, dynamically adjust the collection cycle of perception nodes, edge computing task allocation, and data processing priority. Use real-time collected environmental data to correct the error of the predicted state and feed the corrected state back to the environmental evolution model to complete the closed-loop optimization between perception strategy, situation prediction, and model update.

[0007] Preferably, step S1 specifically includes: Based on the spatial range of the low-altitude area to be monitored, the type of flight activity, and the requirements of low-altitude operation tasks, the target area is spatially divided to form multiple low-altitude sensing areas with independent environmental description capabilities. Multiple types of low-altitude intelligent network sensing nodes are deployed in each sensing area to continuously acquire regional environmental status information through different types of data acquisition equipment. The sensing nodes include flight status sensing nodes, communication environment sensing nodes, micro-meteorological sensing nodes, and space environment sensing nodes. Data from different sources are processed to unify the time scale. Continuously collected data is divided into multiple continuous time windows. Each time window corresponds to the basic time scale for analyzing changes in the low-altitude environment. Data from different sensing nodes within the same time window are aggregated. Based on the spatial distance relationship, a spatial influence weight is established. For the spatial distance between the i-th sensing node and the j-th low-altitude sensing area, the node influence weight is calculated based on the spatial distance, and then the comprehensive environmental status of the area is obtained. Based on regional comprehensive environmental status data, combined with current environmental status, historical status changes, flight behavior change characteristics, and environmental related factors, a low-altitude environmental status primitive is constructed. The environmental status primitive includes environmental status, environmental status changes, regional operational behavior characteristics, and environmental related characteristics. The environmental state primitives generated in different time windows are associated and arranged in chronological order to form a continuous sequence of low-altitude environmental state primitives. The environmental state, change trend, spatial correlation and behavioral characteristics are uniformly encoded to generate an input data structure suitable for machine learning model processing.

[0008] Preferably, step S2 specifically includes: The system analyzes the regional environmental state, state change, operational behavior characteristics, and environmental related factors, transforms the discrete environmental state into a dynamic state description that includes the current state and the direction of change, standardizes the different parameters in the environmental state primitive, constructs a standardized state vector, and introduces a state change direction description parameter to define the direction of environmental state change. A machine learning environmental state evolution model is established by using dynamic environmental state descriptions as input. By learning the historical environmental state change process, the correlation between flight activities, communication quality, weather disturbances and space occupancy factors is explored. The mapping relationship between the current environmental state and the future environmental state is constructed. The dynamic state representation over a continuous time period is input into the evolutionary learning model to generate future state prediction results. The degree of state correlation between different time windows is calculated, and the state correlation degree between adjacent time windows is defined. When the state correlation degree exceeds the set threshold, it means that the current environmental state has not changed significantly relative to the historical state. When the state correlation degree does not exceed the set threshold, it means that the environmental state has changed significantly. The change trend index is calculated in combination with the state change magnitude. The environmental evolution model is corrected by feedback based on the difference between the model prediction results and the actual collected environmental conditions. When the prediction error exceeds the set range, the model parameters and feature weights are adjusted according to the error sources caused by different environmental factors. The training samples are continuously updated using newly added operational data so that the machine learning model can adapt to the low-altitude environmental change patterns under different regions and operating conditions.

[0009] Preferably, step S3 specifically includes: The relationship between environmental state changes between consecutive time windows is analyzed. The current state, future state, state change amount and trend intensity are combined to form environmental state migration units. The similar evolution law of the current environmental change process is analyzed by historical migration pattern matching. By using historical environmental state transition data, statistical analysis is performed on the transformation relationship between different environmental states, and the probability of the current state changing to different future states, i.e., the state transition probability, is calculated. Based on the state transition probability, multiple candidate situation evolution paths are generated, enabling the system to describe the continuous process of the low-altitude environment changing from the current state to the future state. Analyze the environmental impact relationships between different low-altitude sensing areas, combine the spatial distance influence weight with the state migration law to calculate the degree of propagation of environmental changes between areas, and form a low-altitude situation evolution trajectory covering multiple areas; The credibility of the generated situation evolution trajectory is evaluated by comprehensively considering the matching degree of state transition, spatial propagation reliability and prediction error. When the trajectory credibility is lower than the set threshold, the evolution trajectory is dynamically corrected by adjusting model parameters, updating historical samples and enhancing the influence of real-time data.

[0010] Preferably, the process of forming an environmental state transition unit and calculating the state transition probability includes: For the j-th low-altitude sensing area, an environmental state transition unit is constructed between adjacent time windows. This transition unit contains the direction and magnitude of the change from the current environmental state to the next environmental state, and clarifies the direction of change of each environmental factor during the process of the change from the current environmental state to the future state. A migration pattern library is established using historical state migration data. The migration pattern library consists of state migration units formed during historical operation. Each migration pattern includes current state characteristics, state change direction, future state result, and change duration. The current migration unit is matched with the historical migration pattern, and the similarity between the current state migration process and the historical migration pattern is calculated. When the similarity is higher than the set threshold, it is determined that the current environmental change process is consistent with the historical evolution process, and the historical change pattern is used to assist in the subsequent state deduction. Based on the statistical analysis of historical state transition data, the transition relationship between different states is statistically analyzed. The state transition probability is defined as the ratio of the number of times the target state appears after the current state in historical data to the total number of all subsequent states after the current state appears. The evolutionary pattern between different environmental states is obtained by calculating this probability. Based on the magnitude of the transition probability, multiple possible paths are selected to generate the future environmental situation evolution trajectory.

[0011] Preferably, the specific implementation process of analyzing the environmental impact relationships between different low-altitude sensing areas, combining the spatial distance influence weight with the state migration law to calculate the degree of propagation impact of environmental changes between areas, and forming a low-altitude situation evolution trajectory covering multiple areas includes: Regional correlation weights are established based on the spatial distance between different regions. The spatial correlation weights are combined with the state transition probability to calculate the degree of influence of the propagation of situational changes between regions, so as to determine the possible degree of impact of environmental changes in a certain region on the surrounding regions. When environmental changes occur in a certain area, the surrounding affected areas are determined based on spatial correlation weights, thereby increasing the attention paid to future changes in the state of the affected areas and forming a regional-level situation evolution set.

[0012] Preferably, the credibility evaluation of the generated situation evolution trajectory is carried out, comprehensively considering the matching degree of state transition, spatial propagation reliability, and prediction error. When the trajectory credibility is lower than a set threshold, the specific implementation process of dynamically correcting the evolution trajectory by adjusting model parameters, updating historical samples, and enhancing the influence of real-time data includes: The reliability of the trajectory is calculated based on the degree of state transition matching, the degree of spatial propagation stability, and the prediction error. The reliability of the trajectory is determined by a weighted comprehensive analysis of historical state transition matching factors, spatial propagation factors, and prediction error factors. The weights of historical state transition matching factors, spatial propagation factors, and prediction error factors are determined based on historical operating data. A threshold for judging trajectory credibility is set. When the trajectory credibility meets the reliability requirements, the current situation evolution trajectory is directly used for subsequent situation analysis. When the trajectory credibility does not meet the reliability requirements, it is determined that the current trajectory has uncertainty and is dynamically adjusted. The dynamic adjustment methods include increasing the proportion of influence from real-time environmental status data, reducing the influence of abnormal historical samples, re-matching similar historical evolution trajectories, and recalculating the probability of future state migration. After adjustment, the final low-altitude situation evolution trajectory is formed.

[0013] Preferably, step S4 specifically includes: Based on the regional situation evolution trajectory, information on environmental state changes within multiple future time windows is extracted, the amount of change between consecutive states is calculated, and regional active perception demand indicators are constructed by combining state transition probability and trajectory reliability. By setting a preset demand threshold, it is determined whether different regions need to enter an active perception state, and the future environmental change trend is converted into a basis for adjusting perception resources. Based on the regional active perception demand indicators, the collection cycle of perception nodes and edge computing tasks in the low-altitude intelligent network are dynamically adjusted. By normalizing the perception demand of different regions, the proportion of perception resource allocation is determined, and the data update frequency and computing task priority of the corresponding regions are adjusted so that computing resources can serve the regions with future state change needs. The predicted environmental state is corrected in real time by using dynamically adjusted sensing data. The error between the predicted state and the actual collected state is calculated, and the correction method of the prediction result is determined according to the error threshold. The predicted data and real-time data are then integrated to form the corrected environmental state. The environmental state, active perception demand indicators, and trajectory reliability information, after feedback correction, are formed into closed-loop update data and fed back to the environmental evolution model for continuous optimization. By evaluating the accuracy of prediction, resource utilization, and model operation stability, subsequent perception strategies, data collection cycles, and computation task allocation methods are updated.

[0014] Preferably, the specific steps for dynamically adjusting the data collection cycle of sensing nodes and edge computing tasks in the low-altitude intelligent network based on regional active sensing demand indicators include: The future environmental state changes of each region are analyzed by using the situation evolution trajectory, the amount of environmental state change between consecutive prediction time windows is calculated, and the future change intensity index is constructed by combining the state changes of multiple prediction time windows. The time weight of the recent prediction time window is higher than that of the long-term prediction time window. By combining state transition probability and trajectory reliability, a regional future perception demand index is constructed, and a preset perception demand threshold is set. When the regional future perception demand index reaches the active perception condition, the region is designated as a key data update region. The future sensing demand indicators of all low-altitude sensing areas are normalized to convert the sensing demand indicators of different areas into resource allocation criteria under a unified scale. The data collection cycle of the corresponding area is adjusted according to the sensing resource allocation ratio. When the future sensing demand indicator of an area is higher than the threshold, the data collection cycle of that area is shortened. When the future sensing demand indicator of an area is not higher than the threshold, the basic data collection cycle is maintained. The computational task weights for the corresponding regions are determined based on the regional perception requirement indicators. The tasks are then sorted for execution on the edge computing nodes according to the computational task weights. For regions where the computational task weights reach the preset task scheduling conditions, environmental state update tasks, state transition analysis tasks, and situation prediction tasks are executed.

[0015] The technical solution of this invention: A low-altitude intelligent network environment situational awareness system based on machine learning, which is used to execute the aforementioned low-altitude intelligent network environment situational awareness method based on machine learning, comprising: The multi-source heterogeneous environment data acquisition and state primitive construction module is used to collect, synchronize, and fuse multi-source heterogeneous data on flight activities, communication status, environmental disturbances, and space occupancy in the low-altitude intelligent network, and construct a unified expression of environmental state primitives. The Environmental State Evolution Learning and Situation Prediction Module is used to build machine learning prediction models based on environmental state primitives, explore the changing patterns of low-altitude environmental states, analyze the correlation between different environmental parameters, and generate the future state evolution results of the region. The situation evolution trajectory generation and state transition analysis module is used to establish regional situation evolution trajectories based on historical state information and prediction results, calculate environmental state transition relationships and trajectory reliability, and form a description of low-altitude environmental situation changes for future time windows. The active perception and control module is used to calculate the regional perception requirements based on the situation evolution results, dynamically adjust the data acquisition strategy and edge computing task allocation, and use real-time feedback data to correct the prediction results, thereby achieving continuous optimization of the environmental situation perception model and perception strategy.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs a machine learning-based method for environmental situational awareness in low-altitude intelligent networks. By constructing a sequence of environmental state primitives containing spatial attributes, state attributes, and evolutionary trends, it achieves a structured expression and continuous characterization of low-altitude environmental information, providing a high-quality input foundation for machine learning models. Based on machine learning evolutionary models, it deeply mines and dynamically analyzes the coupling relationships between flight activities, communication quality, meteorological disturbances, and space occupancy, significantly improving the accuracy and adaptability of environmental state prediction and effectively addressing the strong dynamism and uncertainty of the low-altitude environment. Furthermore, it generates multi-path situational evolution trajectories through state transition probability calculation, combined with spatial correlation propagation analysis and a dynamic trajectory credibility correction mechanism, enabling the system to comprehensively grasp the direction of environmental changes and quantitatively assess prediction reliability, providing sufficient basis for low-altitude operational decisions. Simultaneously, the constructed active perception optimization mechanism realizes dynamic adjustment of the acquisition strategy and edge computing resources driven by prediction needs, using real-time data feedback to correct prediction deviations and feed back to update the model, forming an efficient closed loop of perception, prediction, and optimization. This enhances the system's adaptability and continuous learning ability to complex and ever-changing low-altitude environments, effectively improving the operational safety and resource utilization efficiency of the low-altitude intelligent network. Attached Figure Description

[0017] Figure 1 This is a flowchart of a low-altitude intelligent network environment situational awareness method based on machine learning proposed in this invention. Figure 2This is an architecture diagram of a low-altitude intelligent network environment situational awareness method system based on machine learning proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes a low-altitude intelligent network environment situational awareness method based on machine learning, which includes the following specific implementation steps: S1. By constructing a low-altitude intelligent network multi-source heterogeneous sensing system, multi-dimensional data such as flight status, communication environment, meteorological changes, and space environment in the low-altitude area are collected synchronously, and unified processing is performed for differences in different data sources, sampling periods, and spatial distribution. On this basis, by combining time window division, spatial influence weight calculation, and environmental change feature extraction methods, the continuously changing low-altitude environmental information is converted into a low-altitude environmental state primitive sequence with spatial attributes, state attributes, and evolutionary trends. S2. Construct a machine learning analysis model for learning the laws of environmental evolution. By dynamically analyzing the current state, direction of change, operational behavior and related environmental factors in the environmental state primitives, a state representation that can describe the process of low-altitude environmental change is formed. Further, the machine learning model is used to learn the coupling relationship between flight activities, communication status, meteorological disturbances and spatial changes, establish the environmental state evolution mapping relationship, and combine trend calculation and feedback correction mechanism to complete the prediction of the future direction of low-altitude environmental change and model adaptive optimization. S3. Construct the low-altitude environmental situation evolution trajectory. By extracting the relationship of environmental state changes within a continuous time window, construct state transition units, and use historical state transition patterns to calculate the probability of environmental state transitions, generating future multi-path situation evolution trajectories. At the same time, combine the spatial correlation between different low-altitude areas to analyze the propagation impact of environmental changes, and optimize the prediction results through trajectory credibility evaluation and feedback correction mechanisms to complete the dynamic extrapolation of the future change process of the low-altitude environment. S4. Construct a future situation-driven proactive perception optimization mechanism. By analyzing the future environmental state change trends in different regions, calculate the regional proactive perception demand indicators, and dynamically adjust the collection cycle of perception nodes, edge computing task allocation, and data processing priority based on the demand results. At the same time, use real-time collected environmental data to correct the error of the predicted state, and feed the corrected state back to the environmental evolution model to complete the closed-loop optimization between perception strategy, situation prediction, and model update.

[0019] In a specific embodiment, step S1 addresses the problems of scattered environmental information sources, inconsistent data collection cycles of different sensing devices, continuous environmental changes, and the inability of traditional data fusion methods to accurately represent changes in the low-altitude operational status during the operation of the low-altitude intelligent network. It proposes a multi-source heterogeneous data organization method for low-altitude environmental evolution analysis. This method focuses on changes in the low-altitude operational environment, unifies the description of multi-dimensional environmental factors affecting the stability of low-altitude operations, and further constructs low-altitude environmental state primitives that reflect the regional environmental status and changing trends. The specific implementation process includes: S11. By constructing a low-altitude intelligent network multi-source heterogeneous sensing system, multi-dimensional data such as flight status, communication environment, meteorological changes, and space environment in the low-altitude area are collected synchronously. Unified processing is performed for different data sources, sampling periods, and spatial distribution differences. Combining time window division, spatial influence weight calculation, and environmental change feature extraction methods, continuously changing low-altitude environmental information is converted into a sequence of environmental state primitives with spatial attributes, state attributes, and evolutionary trends. Specifically: Based on the spatial range of the low-altitude area to be monitored, the type of flight activity, and the requirements of low-altitude operation tasks, the target area is spatially divided to form multiple low-altitude sensing areas with independent environmental description capabilities. The low-altitude sensing area is not simply divided according to a fixed geographical grid, but is determined by combining the range of low-altitude flight activities, communication coverage, and the range of environmental changes, so that each sensing area can correspond to the environmental change unit in the actual low-altitude operation process. For example, in urban low-altitude logistics delivery scenarios, different sensing areas can be divided according to densely built areas, air route intersection areas, and take-off and landing areas; in power line inspection scenarios, corresponding sensing areas can be divided according to the airspace range around transmission lines. After the area is divided, multiple types of low-altitude intelligent network sensing nodes are deployed in each sensing area to continuously acquire regional environmental status information through different types of data acquisition devices; The sensing nodes include at least: (1) Flight status sensing node, used to collect the operating status information of flight equipment entering the low-altitude area, including the number of flight targets, spatial position, altitude, speed, heading angle and motion changes; Among them, the number of flying targets is used to describe the current airspace occupancy level of the area; spatial location is used to determine the area where flight activities occur; speed, altitude, and heading are used to reflect the characteristics of changes in flight behavior. (2) Communication environment sensing node, used to collect the operating status of low-altitude communication network, including communication signal strength, communication delay, data packet loss rate and link connection stability; Since low-altitude flight equipment typically relies on wireless communication to transmit control commands, changes in communication status directly affect the safety of low-altitude operations. (3) Micro-meteorological sensing nodes are used to collect information on changes in low-altitude meteorological environment, including wind speed, wind direction, temperature, humidity, air pressure and local airflow disturbance information; Among them, wind speed changes and airflow disturbances affect the stability of low-altitude flight, and are therefore important factors affecting environmental conditions. (4) Space environment sensing nodes are used to collect information on changes in low-altitude space structure, including the area affected by fixed buildings, the occurrence of temporary obstacles, and changes in airspace occupancy. The data collected by each sensing node is represented as follows: ; In the formula, This represents the set of raw data collected by the i-th sensing node at time t. This represents the m-th type of environmental parameter collected by the i-th sensing node; M represents the number of data categories collected by the sensing node. It should be noted that the above parameters are derived from the actual deployed sensing devices; for example: the flight status node corresponds to the position, altitude, and speed parameters; the communication node corresponds to the signal strength and latency parameters; and the meteorological node corresponds to the wind speed, temperature, and humidity parameters. Furthermore, to ensure that data from different sources can be correlated within a unified spatial range, it is necessary to record the location attributes of each sensing node and define the spatial location of the sensing node: ; In the formula, This represents the horizontal coordinate of the i-th sensing node in the horizontal coordinate system; This represents the vertical coordinate of the i-th sensing node in the horizontal coordinate system; This represents the height coordinates of the i-th sensing node; S12. To address the issues of inconsistent sampling periods and varying spatial influence ranges among different sensing devices, a unified time window is established for the collected data. Influence weights are calculated based on the spatial distance between sensing nodes and the target area. Data from different sources is then mapped to corresponding low-altitude regions to form regional-level comprehensive environmental status data. Specifically: Data from different sources are processed to unify the time scale. Specifically, continuously collected data is divided into multiple continuous time windows, each corresponding to the basic time scale for analyzing changes in low-altitude environmental conditions; the k-th time window is represented as: ; In the formula, This represents the k-th time window; Indicates the start time of the time window; Indicates the end time of the time window; Data from different sensing nodes within the same time window are aggregated: ; In the formula, This represents the set of all perceived data within the k-th time window; By using time window processing, data with different sampling frequencies can be converted to a unified time scale; Furthermore, since different sensing nodes have different degrees of influence on the environmental state of the same area, this embodiment establishes spatial influence weights based on spatial distance relationships; The spatial distance between the i-th sensing node and the j-th low-altitude sensing area: ; In the formula, Indicates the coordinates of the center of the target perception area; Calculate the node influence weight based on spatial distance: ; In the formula, The weight represents the influence of the i-th node on the environmental state of the j-th region; N represents the number of sensing nodes participating in the calculation of the region state. Further information on the overall environmental status of the region is available: ; In the formula, This represents the overall environmental state of the j-th region within the k-th time window; S13. Based on comprehensive regional environmental status data, combined with current environmental status, historical changes, flight behavior characteristics, and related environmental factors, low-altitude environmental status primitives are constructed. Traditional static environmental parameters are transformed into dynamic descriptive units containing status information and trends, enabling each environmental status primitive to reflect the real-time status and evolution direction of the low-altitude environment within a specific region. Specifically: After obtaining the overall regional environmental status, environmental status primitives are further constructed; Traditional low-altitude sensing methods mainly focus on current environmental parameters, such as the current number of flights and the current communication quality. However, low-altitude environmental risks are often not determined by the state at a single moment, but are formed by the continuous process of environmental change. Therefore, by introducing environmental change characteristics, the current state and the trend of change are jointly encoded to form environmental state primitives that can describe the evolution characteristics of the low-altitude environment; For the j-th region, generate the environment state primitive within the time window k: ; In the formula, This represents the environmental state primitive corresponding to region j within time window k; It represents the amount of change in the environmental state, that is, the degree of change in the current environmental state relative to the historical state; Indicates the characteristics of regional operational behavior; Indicates environmental association characteristics; The environmental change was calculated as follows: ; In the formula, Indicates the environmental state of the previous time window; Regional operational characteristics include: the rate of change in the number of flying targets; the concentration of flight paths; and the increasing trend of airspace occupancy. Environmental factors include: changes in communication quality; changes in weather disturbances; and changes in spatial barriers. S14. Arrange the environmental state primitives generated in different time windows according to time order to form a continuous low-altitude environmental state primitive sequence, and uniformly encode the environmental state, change trend, spatial correlation and behavioral characteristics in it to generate an input data structure suitable for machine learning model processing, specifically: Further organizing the environmental state primitives in chronological order, for region j, forms a continuous state sequence: ; In the formula, Represents the sequence of environmental state primitives in region j; K represents the number of historical time windows; Furthermore, the different attributes in the state primitives are standardized and encoded to form machine learning inputs: ; In the formula, This represents the machine learning input vector corresponding to region j; This represents the nth environmental state characteristic; The features include: current environmental state features; environmental change trend features; spatial correlation features; and flight behavior change features. Ultimately, a sequence of low-altitude environmental state primitives is formed.

[0020] In a specific embodiment, step S2, based on the low-altitude environment state primitive sequence constructed in step S1, further establishes a machine learning analysis model oriented towards the environmental evolution process, taking into account the characteristics of the low-altitude environment being highly dynamic, strongly coupled, and having gradual state changes. The continuous environmental state primitives formed in step S1 are used as time series analysis objects. By learning the correlation between different factors in the historical environmental state change process, a model of the low-altitude environment state change pattern is established. Its specific implementation process includes: S21. Further analyze the regional environmental state, state changes, operational behavior characteristics, and related environmental factors to transform the discrete environmental state into a dynamic state description that includes the current state and direction of change. Through standardization of environmental characteristics and calculation of change trends, a dynamic environmental state representation suitable for machine learning model input is formed, specifically: Based on the environmental state primitive sequence output in step S1, the environmental state change information within a continuous time range is further analyzed. Since the environmental state primitive generated in step S1 already contains the regional environmental state, state change amount, operational behavior characteristics and related environmental factors, this embodiment extracts dynamic features that can reflect the low-altitude environmental change pattern from the existing state primitives. Specifically, the different parameters in the environmental state primitive are standardized to construct a standardized state vector: ; In the formula, This represents the standardized environmental state vector of region j within time window k; This represents the result of normalization for the nth environmental state feature; Among them, each characteristic parameter is determined based on the actual collected data. For example, the flight density change rate corresponds to flight behavior characteristics; the communication delay change rate corresponds to communication stability characteristics; the wind speed change rate corresponds to meteorological disturbance characteristics; and the airspace occupancy change rate corresponds to spatial state characteristics. Furthermore, to reflect the continuity of low-altitude environmental changes, a parameter describing the direction of state change is introduced, defining the direction of environmental state change: ; In the formula, This indicates the direction of environmental state change in the j-th region within the time window k; This represents the environmental state vector corresponding to the previous time window; S22. Using a dynamic environmental state description as input, establish a machine learning environmental state evolution model. By learning the historical environmental state change process, explore the correlation between factors such as flight activities, communication quality, weather disturbances, and space occupancy, and construct a mapping relationship between the current environmental state and the future environmental state. Specifically: Further, a machine learning-based environmental state evolution model will be established to learn the patterns of low-altitude environmental state changes. Since low-altitude environmental changes have obvious coupling characteristics, for example: increased flight density may lead to enhanced competition for communication resources; decreased communication quality may lead to adjustments in flight behavior; and enhanced weather disturbances may change the distribution of flight trajectories. Therefore, this embodiment learns the correlation between different environmental factors through a machine learning model. Specifically, the dynamic state representation over a continuous time period is input into the evolutionary learning model: ; In the formula, This represents the hidden state of the environment evolution of region j within time window k; This indicates the historical evolution state of the previous time window; Represents a machine learning state evolution mapping function; Among them, the machine learning state evolution mapping function By training it with historical environmental state sequences, it learns: how current state factors affect future changes; how environmental states transition under different combinations of factors; and whether environmental changes have a continuous trend. It is necessary to further explain the machine learning state evolution mapping function. This function describes the intrinsic mapping relationship of low-altitude environmental state changes over time. Its purpose is to learn the correlation patterns between factors such as flight activities, communication status, meteorological disturbances, and the space environment based on dynamic environmental state information within the current time window and historical evolutionary state information. Through training, it obtains the relationship between the current environmental state and the future evolutionary state. This function can be implemented by a neural network model based on time series learning, an ensemble learning model, or other machine learning structures suitable for state prediction. Its specific structure is determined based on training with historical low-altitude operational data. In this embodiment, it can be implemented using a Long Short-Term Memory (LSTM) network, a Gated Recurrent Unit (GRU) network, or a Transformer time-series coding model based on temporal feature learning. The Transformer model is preferred, as it captures the long-term correlation relationships between multiple factors such as flight activities, communication status, and meteorological changes through a self-attention mechanism. Further generate future state prediction results: ; In the formula, This indicates the predicted environmental state for the next time window; Represents the environmental evolution state mapping function; For example, when the model inputs: an increase in the number of flights in the current area; a decrease in communication quality; and an increase in wind speed fluctuations, the model can determine, based on historical operational patterns, whether the area is likely to continue to experience increased operational pressure in the future. It should be noted that the environmental evolution state mapping function This represents the mapping process that transforms the environmental evolution state generated within a machine learning model into a description of a specific future environmental state. It is primarily used to restore hidden state information into observable environmental parameter changes. This function establishes internal parameter relationships based on historical state evolution patterns, mapping the state features learned by the model into predicted results such as changes in flight activity, communication status, meteorological impact, and space occupancy within a future time window. This enables early prediction of the low-altitude environmental situation. In this embodiment, it can be implemented using a multilayer perceptron (MLP), a fully connected neural network, or a prediction-decoding network based on an attention mechanism. A multilayer perceptron structure is preferred, using nonlinear mapping relationships to convert the evolution state into interpretable environmental indicators such as flight density, communication stability, meteorological disturbance levels, and space occupancy status. S23. Based on the output results of the environmental state evolution model, further analyze the correlation and variation of environmental states between different time windows, extract the migration patterns of low-altitude environmental states, and calculate environmental change trend indicators by combining state continuity and change intensity, thereby achieving a quantitative description of regional environmental stability, continuous change state, and potential change direction. Specifically: The degree of state correlation between different time windows is calculated, and the state correlation degree between adjacent time windows is defined: ; In the formula, This indicates the degree of correlation between the current state and historical states; when When the threshold is exceeded, it indicates that the environmental state is changing slowly; when A decrease indicates a significant change in the environmental state; Further, combine the magnitude of state changes to calculate the trend indicator: This trend indicator can be used to determine whether: the regional environment remains stable; whether the environment is continuously deteriorating; and whether there is a rapid trend of change. In the formula, Indicates the intensity of regional environmental change trends; Indicates the magnitude of the state change; and These represent the adjustment weights for the magnitude of change and the degree of state abrupt change, respectively. in, and This is determined based on the results of historical environmental changes and is used to adjust for the impact of different factors on trend judgment. S24. Based on the difference between the model's prediction results and the actual environmental conditions, the environmental evolution model is corrected through feedback. When the prediction error exceeds a set range, the model parameters and feature weights are adjusted according to the error sources caused by different environmental factors, and the training samples are continuously updated using newly added operational data. This enables the machine learning model to adapt to the changing patterns of the low-altitude environment under different regions and operating conditions. Specifically: The machine learning model is updated using actual operational feedback data, and the model's predictions are compared with the actual collected data. ; In the formula, This represents the prediction error for the j-th region; This indicates that the model predicts the state of the environment; When the prediction error exceeds the preset threshold, it indicates that the current model's adaptability to the environmental change patterns in the region has decreased; at this time, the model parameters should be adjusted according to the source of the error. For example, if the long-term prediction of changes in communication status has a large deviation, the weight of communication-related features should be increased; if changes in weather conditions reduce the accuracy of predictions, the proportion of weather status samples should be increased. Through continuous feedback, the model can automatically adjust according to changes in the actual operating environment.

[0021] In a specific embodiment, step S3 extracts the state change correlation from the output of the environmental evolution model to form environmental state migration units. Based on historical environmental change patterns, it calculates the migration probability between different states, generates future environmental state evolution paths, and further analyzes the propagation process of environmental changes between different regions by combining the spatial influence relationships between low-altitude areas. Based on the prediction feedback results, it evaluates the credibility of the generated situational evolution trajectory and dynamically corrects it. The specific implementation process includes: S31. Analyze the relationship between environmental state changes between consecutive time windows, combine the current state, future state, state change amount, and trend intensity to form environmental state migration units, and analyze the similar evolutionary patterns of the current environmental change process through historical migration pattern matching analysis, specifically: Based on the hidden environmental evolution states, environmental trend indicators, and future state prediction results output in step S2, the relationship between environmental state changes within a continuous time window is analyzed. Further analysis of the environmental state change patterns between adjacent time windows is conducted, and a single environmental change process is encapsulated as a state transition unit. Specifically, for the j-th low-altitude sensing area, an environmental state transition unit is constructed between time window k and time window k+1. ; In the formula, This represents an environmental state transition process in the j-th region during the k-th time window; This indicates the direction and magnitude of the change in environmental state from the current moment to the next moment; The change in environmental state is further expressed as follows: ; Identify the direction of change of each environmental factor as the current environmental state changes towards the future state; For example, when the flight density characteristic increases from 0.45 to 0.65, it corresponds to an increasing trend in flight activity; when the communication stability characteristic decreases from 0.8 to 0.5, it corresponds to a deteriorating trend in the communication environment; and when the weather disturbance characteristic continues to increase, it corresponds to a decreasing trend in environmental stability. Furthermore, in order to determine whether the current environmental change process is similar to the historical change process, this embodiment uses historical state transition data to establish a transition pattern library. The transition pattern library consists of a large number of state transition units formed during historical operation. Each transition pattern includes: current state characteristics; direction of state change; future state result; duration of change. Match the current migration unit with historical migration patterns: ; In the formula, This indicates the degree of similarity between the current state transition process and historical transition patterns; This represents the historical h-th type of migration pattern; Indicates the magnitude of the corresponding vector; when When the value exceeds the set threshold, it indicates that the current environmental change process is highly consistent with a certain type of historical evolution process, and this historical change pattern can be used to assist in the subsequent state prediction. S32. Using historical environmental state transition data, statistical analysis is performed on the transition relationships between different environmental states to calculate the probability of the current state transitioning to different future states. Based on the transition probabilities, multiple candidate situation evolution paths are generated, enabling the system to describe the continuous process of low-altitude environment changes from the current state to future states. Specifically: After obtaining the environmental state transition unit, the possible future state transition direction is calculated based on the historical state change pattern and the current environmental state. Since the low-altitude environment is affected by multiple factors, the same current state may correspond to multiple future change results. For example, when the number of flights in the area increases, it may remain stable in the future, or it may further enter a high-pressure state due to the increase in communication load. Therefore, this embodiment generates multiple candidate evolution paths based on the probability of state transitions, and defines the state transition probability by statistically analyzing the transition relationships between different states based on historical state transition data: ; In the formula, This represents the probability that the j-th region will transition from state a to state b, i.e., the state transition probability. This indicates the number of times state a subsequently occurs as state b in historical data; This represents the number of all subsequent states after state a occurs; This probability calculation allows us to obtain the evolutionary patterns between different environmental states; For example: For a certain city's low-altitude area: Current state: "Medium flight density + stable communication"; Historical data statistics: The probability of converting to "Increased flight density + Increased communication load" is 0.6; the probability of maintaining a stable state is 0.3; the probability of returning to a low-load state is 0.1. Furthermore, the trajectory of future environmental states is generated based on migration probabilities: ; In the formula, This represents the future environmental evolution trajectory of the j-th region; Indicates the current environmental state; It represents the predicted state for a continuous future time period, which includes normalized multidimensional environmental state parameters, such as: parameters on changes in the number of flight activities, parameters on regional space occupancy, parameters on communication resource usage, parameters on environmental disturbances, and parameters on flight operation stability; m represents the number of prediction time steps; During trajectory generation, multiple possible paths are selected based on the migration probability and sorted according to their probability. For example: Path A: Stable state → Slight change → High load state; Path B: Stable state → Communication fluctuation → State recovery; S33. Analyze the environmental impact relationships between different low-altitude sensing areas, combine the spatial distance influence weight with the state transition law, calculate the degree of propagation of environmental changes between areas, and form a low-altitude situation evolution trajectory covering multiple areas, specifically: Because the low-altitude environment has obvious spatial correlation characteristics, different areas are not independent of each other; for example, increased flight activity around an airport will affect surrounding route resources; decreased communication quality in densely built-up areas may affect the communication load of neighboring areas. Therefore, this embodiment further utilizes the spatial correlation established in step S1 to analyze the propagation process of environmental conditions between different regions; and establishes regional correlation weights based on the spatial distance between different regions (the closer the distance, the more significant the impact of environmental changes, and the higher the corresponding weight): ; In the formula, This represents the spatial influence weight of region u on region v; v represents the spatial distance between two regions; v represents other regions that are spatially related to region u. Furthermore, spatial association weights are combined with state transition probabilities: ; In the formula, Indicates the degree of propagation and impact of changes in the situation between regions; Indicates the probability of state transition in a region; This calculation can determine the extent to which environmental changes in a certain area may affect the surrounding areas. For example: Region A: Rapid increase in flight volume; Decrease in communication resources; Based on spatial correlation: Region B is relatively close; Historical data shows that changes in A often affect B; Therefore, we should increase our attention to future changes in the state of Region B in advance. Ultimately, a regional-level situational evolution set is formed: ; In the formula, G represents the set of overall low-altitude regional situation evolution trajectories; This represents the evolutionary trajectory corresponding to the nth region; S34. Evaluate the credibility of the generated situation evolution trajectory, comprehensively considering the matching degree of state transition, spatial propagation reliability, and prediction error. When the trajectory credibility is insufficient, dynamically correct the evolution trajectory by adjusting model parameters, updating historical samples, and enhancing the influence of real-time data. Specifically: The generated situation trajectory is dynamically corrected based on the reliability of state transition, the stability of spatial propagation, and the prediction error, and the trajectory reliability is calculated: ; And define the prediction error: ; In the formula, Indicates the credibility of the situation evolution trajectory of the j-th region; This indicates the degree of spatial propagation stability, that is, the degree of propagation impact of inter-regional situational changes in step S33. ; Indicates prediction error; Indicates the weight of historical state migration matching factors; Indicates the weight of spatial propagation factors; Indicates the weight of prediction error factors; This represents the i-th feature value in the predicted state of the j-th region; This represents the i-th feature value in the actual data acquisition state; It should be noted that the weight parameters , and Determined based on historical operational data; Specifically: when the historical environmental change patterns of a certain low-altitude region are relatively stable, improve... The proportion of [something] allows the laws of historical evolution to play a greater role; when inter-regional influence relationships are significant, [something] should be increased. The proportion; when the quality of real-time sensing data is high, improve The constraint effect on the error term; Further set the credibility judgment threshold: (Preset trajectory credibility threshold); when This indicates that the current situation evolution trajectory meets the reliability requirements and can be directly used for subsequent situation analysis; when This indicates that the current trajectory has significant uncertainty and requires dynamic adjustment; Adjustment methods include, but are not limited to: increasing the proportion of influence from real-time environmental state data; reducing the influence of abnormal historical samples; re-matching similar historical evolution trajectories; and recalculating the probability of future state transitions. After adjustments, the final low-altitude situation evolution trajectory was formed.

[0022] In a specific embodiment, step S4 calculates future environmental change requirements based on the regional situation evolution trajectory output in step S3, forming a regional active sensing requirement evaluation result; and adjusts the data acquisition cycle of sensing nodes and the edge computing task allocation method based on the sensing requirement evaluation result; further utilizes real-time acquired data to perform feedback correction on the predicted state, reducing model prediction bias; and re-inputs the corrected environmental state into the environmental evolution model to achieve closed-loop optimization between sensing strategy, state prediction, and model update. The specific implementation process includes: S41. Based on the regional situational evolution trajectory output in step S3, extract environmental state change information within multiple future time windows, calculate the change between consecutive states, and construct regional active sensing demand indicators by combining state transition probability and trajectory reliability. Use preset demand thresholds to determine whether different regions need to enter an active sensing state, converting future environmental change trends into a basis for adjusting sensing resources. Specifically: The aforementioned situational evolution trajectory is used to analyze the future environmental state changes in each region, and the data collection requirements for different regions are determined based on the changing trends. Specifically, the state change between consecutive prediction time windows is further calculated: ; In the formula, This represents the change in environmental state of the j-th region from the (r-1)-th prediction time window to the r-th prediction time window; This indicates the current environmental state within the forecast time window; This indicates the environmental state of the previous prediction time window; By further integrating the state changes across multiple forecast time windows, an index for the intensity of future changes is constructed: ; In the formula, This represents the intensity of future environmental state changes in the j-th region; This represents the time weight corresponding to the r-th prediction time window; This indicates the degree of change in environmental state after normalization. Among them, time weight The settings are based on how close the predicted time is to the current time window, and are used to reflect the degree of impact of recent environmental changes on the current perception adjustment; Further combining the state transition probabilities output in step S3 and trajectory credibility Construct regional future perception demand indicators: ; In the formula, This represents the future perceived demand index for the j-th region; Further set preset perception demand thresholds, based on Determine whether the area has entered an active sensing state; In the formula, This indicates a pre-set threshold for judging perception needs, which is determined based on the data update needs corresponding to changes in different environmental states in historical operational data. When a region meets the conditions for active sensing, the system will designate that region as a key data update region and increase the data acquisition frequency of the corresponding sensing nodes and the edge computing processing priority in subsequent steps. S42. Based on the regional proactive sensing demand indicators, dynamically adjust the data acquisition cycle and edge computing tasks of sensing nodes in the low-altitude intelligent network. By normalizing the sensing demands of different regions, determine the allocation ratio of sensing resources, and adjust the data update frequency and computing task priority of corresponding regions, so that computing resources prioritize serving regions with future state change demands. Specifically: Furthermore, based on the perception needs indicators of different regions, the data acquisition process of the perception nodes and the task processing process of the edge computing nodes in the low-altitude intelligent network will be dynamically adjusted. In the actual deployment of low-altitude intelligent networks, the number of edge sensing nodes, communication bandwidth, and computing resources are all limited, making it impossible to infinitely increase the data acquisition frequency and computing processing frequency in all areas. Therefore, it is necessary to rationally allocate the limited data acquisition and computing resources according to the needs of future environmental changes. This step utilizes the regional future perception demand index calculated in step S41. Establish a mapping relationship between sensing needs and data acquisition strategies, so that sensing nodes can dynamically adjust the data acquisition method according to future environmental change trends; The future sensing demand indicators for all low-altitude sensing areas are normalized to convert the sensing demand indicators of different areas into a resource allocation basis under a unified scale. ; In the formula, This represents the proportion of sensing resources allocated to the j-th region. This indicates the total number of sensing areas currently defined by the Low-Altitude Intelligent Network; Furthermore, the data collection cycle for the corresponding area is adjusted according to the allocation ratio of sensing resources: ; In the formula, This represents the adjusted data collection period for the j-th region; This indicates the initial data acquisition cycle set by the system; Among them, the initial data acquisition cycle set by the system Pre-set according to the hardware sampling capability of the sensing node, the transmission capability of the communication link, and the requirements for updating the status of the low-altitude environment; When the region's future perceived demand indicators meet This indicates that there is a high probability of a change in the future environmental state of the area, which can shorten the data collection cycle of the area and enable edge nodes to obtain more continuous environmental state information. When the region's future perceived demand indicators meet At the same time, maintain the basic data collection cycle to ensure that the sensing nodes remain in normal operating condition; For example, in the scenario of urban low-altitude logistics operation, if the flight activity status of a certain area is continuously changing within multiple time windows according to the prediction results of step S3, and the state transition probability meets the preset conditions, then the system will automatically reduce the collection cycle of that area, so that the number of flight activities, communication resource occupation and spatial status changes can be obtained more timely. Furthermore, to avoid uneven task load on edge computing nodes after data collection adjustments, this embodiment also adjusts the edge computing task queue according to regional perception demand indicators. Specifically, based on: Determine the computation task weight corresponding to the j-th region; In the formula, represents the weight of the computation task corresponding to the j-th region; W represents the total computing resources currently available for allocation to the edge computing node. Based on the weight of the computational tasks, the tasks executed on the edge computing nodes are sorted: When the preset task scheduling conditions are met, the following tasks in this area will be executed first: environmental status update task; state transition analysis task; situation prediction task. For regions where the calculated task weight does not meet the task scheduling conditions, data processing is performed according to the basic task queue. S43. Using the dynamically adjusted sensing data, the environmental state predicted in step S3 is corrected in real time. By calculating the error between the predicted state and the actual collected state, and determining the correction method based on the error threshold, the predicted data and real-time data are fused to form the corrected environmental state, specifically: After step S42, each sensing node of the low-altitude intelligent network continuously acquires environmental data according to the adjusted acquisition strategy and inputs the real-time acquisition results into the environmental situational awareness model. Because the low-altitude environment is dynamically changing, with factors such as increased temporary flight missions, changes in communication environments, weather conditions, and space occupancy, there is a certain deviation between the future situation evolution trajectory generated in step S3 and the actual environmental state. If the original prediction results are directly used to adjust the perception strategy for a long period of time, errors may gradually accumulate, causing subsequent perception resource allocation to deviate from the actual environmental requirements. Therefore, this embodiment uses real-time acquired data to provide feedback correction for the predicted environmental state. Let the predicted state of the j-th region corresponding to time window k be... The environmental status obtained in real time is ; Obtain the error between the predicted state and the actual state. Further set the prediction error threshold (The maximum allowable prediction error range of the system is determined based on the error statistics between historical prediction results and actual operating results). when If the current prediction results meet the requirements of environmental state analysis, the original situation evolution model output should be maintained. when This indicates a discrepancy between the current environmental state change process and historical learning patterns, necessitating an increase in the proportion of real-time data in the state correction process. Further calculation of the real-time feedback ratio: ; In the formula, Indicates the real-time feedback correction ratio; This indicates the maximum allowable prediction error of the system; Based on the feedback ratio, the predicted state and the real-time state are fused, enabling the model to automatically adjust the prediction results according to changes in the operating environment. ; In the formula, This indicates the environmental state after feedback and correction. For example, when the flight density in a certain area changes rapidly due to temporary flight missions, a deviation occurs between the real-time collected status and the predicted status. Based on the error calculation results, the proportion of the real-time status is increased so that the corrected environmental status is more consistent with the current operating conditions. S44. The environmental state, active sensing demand indicators, and trajectory reliability information, after feedback correction, are formed into closed-loop update data, which is then fed back to the environmental evolution model for continuous optimization. By evaluating the prediction accuracy, resource utilization, and model stability, subsequent sensing strategies, data collection cycles, and computational task allocation methods are updated. Specifically: Using the feedback correction state output in step S43, the environmental evolution model and perception control strategy are continuously updated, enabling the system to continuously adjust its processing capabilities based on new environmental data accumulated during long-term operation. Specifically, this embodiment integrates the environmental state after feedback correction, regional perception demand indicators, and situation evolution credibility to form a new environmental state update sample; and adds the updated sample to the historical environmental state set to update the state transition relationship and prediction parameters in the machine learning model; further, based on the prediction effect after the model update, the subsequent perception resource allocation strategy is optimized. The feedback-corrected environmental state calculated in step S43 is taken as the effective environmental state of the current time window. For the j-th region, the closed-loop update state corresponding to time window k is represented as follows: ; In the formula, This represents the closed-loop update state formed by the j-th region in the k-th time window; Furthermore, the closed-loop update states formed by multiple consecutive time windows are combined into an online update sample set: ; In the formula, This represents the set of environmental update samples for the entire region corresponding to the k-th time window; After obtaining new environmental update samples, the environmental evolution model is further adjusted based on actual operational feedback. Since the machine learning model in step S2 has already established a mapping relationship between the environmental state and the future state, this embodiment, based on the original model, corrects the state transition parameters according to the newly added operational data, so that the model gradually adapts to the current low-altitude operating environment. Specifically, the model prediction error change index is defined as follows: ; In the formula, This indicates the change in prediction error between adjacent time windows; This indicates the model prediction error for the current time window; This indicates the model prediction error in the previous time window; when If the current model prediction error shows an increasing trend, then reduce the influence ratio of historical state transition parameters and increase the training weight of the latest feedback sample. when If the model prediction error does not increase, then the current model parameter update strategy should be maintained. Further define the feedback sample update weights: ; In the formula, This indicates the updated weights of the feedback samples within the current time window; This represents the basic update weights, which are determined based on the stability of the data during the initial training phase of the model. This represents the error adjustment coefficient, which is determined based on the model's response speed to environmental changes. Further evaluate the closed-loop optimization effect, adjust subsequent sensing strategies based on the evaluation results, and construct a comprehensive evaluation index: ; In the formula, O represents the current closed-loop perception optimization evaluation index; Indicators representing the accuracy of environmental state predictions; Indicators representing perceived resource utilization; Indicates the stability index of the model operation; , and These represent the weights of the corresponding evaluation indicators; Among them, the accuracy index of environmental status prediction Calculated based on prediction error: Perceived resource utilization indicators Determined based on the number of effective environmental status updates generated per unit of computing resources, used to evaluate the resource utilization of edge computing nodes; model operation stability index Calculated based on the magnitude of model parameter changes over multiple consecutive time windows, to avoid over-adjustment of model parameters due to a single abnormal data point; Further optimize the evaluation threshold ,when When, maintain the current perception strategy; when In this case, adjustments will be made based on the evaluation results: regional perception priority, data collection cycle, edge computing task allocation ratio, and model update frequency. For example: when the prediction accuracy index declines, increase the proportion of real-time environmental status data involved; when the perception resource utilization index declines, recalculate the regional perception demand index; when the model stability index declines, reduce the impact of short-term abnormal data on model parameter updates. Through the above adjustments, the environmental situational awareness process of the low-altitude intelligent network can be continuously optimized based on long-term operational results.

[0023] Example 2, as Figure 2As shown, the present invention proposes a low-altitude intelligent network environment situational awareness system based on machine learning, which is used to execute a low-altitude intelligent network environment situational awareness method based on machine learning proposed in Embodiment 1. It includes: a multi-source heterogeneous environment data acquisition and state primitive construction module, an environment state evolution learning and situation prediction module, a situation evolution trajectory generation and state transition analysis module, and an active perception and control module.

[0024] The multi-source heterogeneous environment data acquisition and state primitive construction module is used to collect, synchronize, and fuse features of multi-source heterogeneous data such as flight activities, communication status, environmental disturbances, and space occupancy in the low-altitude intelligent network, and construct a unified expression of environmental state primitives. The Environmental State Evolution Learning and Situation Prediction Module is used to build machine learning prediction models based on environmental state primitives, explore the changing patterns of low-altitude environmental states, analyze the correlation between different environmental parameters, and generate the future state evolution results of the region. The situation evolution trajectory generation and state transition analysis module is used to establish regional situation evolution trajectories based on historical state information and prediction results, calculate environmental state transition relationships and trajectory reliability, and form a description of low-altitude environmental situation changes for future time windows. The active perception and control module is used to calculate the regional perception requirements based on the situation evolution results, dynamically adjust the data acquisition strategy and edge computing task allocation, and use real-time feedback data to correct the prediction results, thereby achieving continuous optimization of the environmental situation perception model and perception strategy.

[0025] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A machine learning-based method for environmental situational awareness in low-altitude intelligent networks, characterized in that, The specific implementation steps include the following: S1. Construct a low-altitude intelligent network multi-source heterogeneous sensing system to synchronously collect multi-dimensional data and uniformly process differences in data sources, sampling periods and spatial distribution. Combine time window division, spatial influence weight calculation and environmental change feature extraction to convert continuously changing low-altitude environmental information into a sequence of environmental state primitives with spatial attributes, state attributes and evolutionary trends. Specifically, step S1 includes: Based on the spatial range of the low-altitude area to be monitored, the type of flight activity, and the requirements of low-altitude operation tasks, the target area is spatially divided to form multiple low-altitude sensing areas with independent environmental description capabilities. Multiple types of low-altitude intelligent network sensing nodes are deployed in each sensing area to continuously acquire regional environmental status information through different types of data acquisition equipment. The sensing nodes include flight status sensing nodes, communication environment sensing nodes, micro-meteorological sensing nodes, and space environment sensing nodes. Data from different sources are processed to unify the time scale. Continuously collected data is divided into multiple continuous time windows. Each time window corresponds to the basic time scale for analyzing changes in the low-altitude environment. Data from different sensing nodes within the same time window are aggregated. Based on the spatial distance relationship, a spatial influence weight is established. For the spatial distance between the i-th sensing node and the j-th low-altitude sensing area, the node influence weight is calculated based on the spatial distance, and then the comprehensive environmental status of the area is obtained. Based on regional comprehensive environmental status data, combined with current environmental status, historical status changes, flight behavior change characteristics, and environmental related factors, a low-altitude environmental status primitive is constructed. The environmental status primitive includes environmental status, environmental status changes, regional operational behavior characteristics, and environmental related characteristics. The environmental state primitives generated in different time windows are associated and arranged in chronological order to form a continuous sequence of low-altitude environmental state primitives. The environmental state, change trend, spatial correlation and behavioral characteristics are uniformly encoded to generate an input data structure suitable for machine learning analysis model processing. S2. Construct a machine learning analysis model for learning the laws of environmental evolution. By dynamically analyzing the current state, direction of change, operational behavior and environmental related factors in the environmental state primitives, a state representation that can describe the process of low-altitude environmental change is formed. The machine learning analysis model is used to learn the coupling relationship between flight activities, communication status, meteorological disturbances and spatial changes, establish the environmental state evolution mapping relationship, and combine trend calculation and feedback correction to complete the prediction of the future direction of low-altitude environmental change and model adaptive optimization. Specifically, step S2 includes: The system analyzes the regional environmental state, state change, regional operational behavior characteristics, and environmental related factors, transforms the discrete environmental state into a dynamic environmental state description that includes the current state and the direction of change, standardizes the different parameters in the environmental state primitive, constructs a standardized state vector, and introduces a state change direction description parameter to define the direction of environmental state change. A machine learning environmental state evolution model is established by using dynamic environmental state descriptions as input. By learning the historical environmental state change process, the correlation between flight activities, communication quality, weather disturbances and space occupancy factors is explored. The mapping relationship between the current environmental state and the future environmental state is constructed. The dynamic environmental state descriptions over a continuous period of time are input into the machine learning environmental state evolution model to generate future state prediction results. The degree of state correlation between different time windows is calculated, and the state correlation degree between adjacent time windows is defined. When the state correlation degree exceeds the set threshold, it means that the current environmental state has not changed significantly relative to the historical state. When the state correlation degree does not exceed the set threshold, it means that the environmental state has changed significantly. The change trend index is calculated in combination with the state change magnitude. The machine learning environment state evolution model is corrected by feedback based on the difference between the model prediction results and the actual collected environment state. When the prediction error exceeds the set range, the model parameters and feature weights are adjusted according to the error sources caused by different environmental factors. The training samples are continuously updated using newly added operating data so that the machine learning analysis model can adapt to the low-altitude environment change patterns under different regions and operating conditions. S3. Construct the low-altitude environment situation evolution trajectory. By extracting the relationship of environmental state changes within a continuous time window, construct state migration units, and use historical state migration patterns to calculate the probability of environmental state migration, generate future multi-path situation evolution trajectories, analyze the impact of environmental change propagation by combining the spatial correlation between different low-altitude areas, and optimize the prediction results through trajectory credibility evaluation and feedback correction, to complete the dynamic extrapolation of the future change process of the low-altitude environment. S4. Construct a future situation-driven proactive perception optimization mechanism. Calculate regional proactive perception demand indicators by analyzing the future environmental state change trends of different regions. Based on the demand results, dynamically adjust the collection cycle of perception nodes, edge computing task allocation, and data processing priority. Use real-time collected environmental data to correct the error of the predicted state and feed the corrected state back to the environmental evolution model to complete the closed-loop optimization between perception strategy, situation prediction, and model update.

2. The method for low-altitude intelligent network environment situational awareness based on machine learning according to claim 1, characterized in that, Step S3 specifically includes: The relationship between environmental state changes between consecutive time windows is analyzed. The current state, future state, state change amount and trend intensity are combined to form environmental state migration units. The similar evolution law of the current environmental change process is analyzed by historical migration pattern matching. By using historical environmental state transition data, statistical analysis is performed on the transformation relationship between different environmental states, and the probability of the current state changing to different future states, i.e., the state transition probability, is calculated. Based on the state transition probability, multiple candidate situation evolution paths are generated, enabling the system to describe the continuous process of the low-altitude environment changing from the current state to the future state. Analyze the environmental impact relationships between different low-altitude sensing areas, combine the spatial distance influence weight with the state migration law to calculate the degree of propagation of environmental changes between areas, and form a low-altitude situation evolution trajectory covering multiple areas; The credibility of the generated situation evolution trajectory is evaluated by comprehensively considering the matching degree of state transition, spatial propagation reliability and prediction error. When the trajectory credibility is lower than the set threshold, the evolution trajectory is dynamically corrected by adjusting model parameters, updating historical samples and enhancing the influence of real-time data.

3. The method for low-altitude intelligent network environment situational awareness based on machine learning according to claim 2, characterized in that, The process of forming environmental state transition units and calculating state transition probabilities includes: For the j-th low-altitude sensing area, an environmental state transition unit is constructed between adjacent time windows. This transition unit contains the direction and magnitude of the change from the current environmental state to the next environmental state, and clarifies the direction of change of each environmental factor during the process of the change from the current environmental state to the future state. A migration pattern library is established using historical state migration data. The migration pattern library consists of state migration units formed during historical operation. Each migration pattern includes current state characteristics, state change direction, future state result, and change duration. The current migration unit is matched with the historical migration pattern, and the similarity between the current state migration process and the historical migration pattern is calculated. When the similarity is higher than the set threshold, it is determined that the current environmental change process is consistent with the historical evolution process, and the historical change pattern is used to assist in the subsequent state deduction. Based on the statistical analysis of historical state transition data, the transition relationship between different states is statistically analyzed. The state transition probability is defined as the ratio of the number of times the target state appears after the current state in historical data to the total number of all subsequent states after the current state appears. The evolutionary pattern between different environmental states is obtained by calculating this probability. Based on the magnitude of the transition probability, multiple possible paths are selected to generate the future environmental situation evolution trajectory.

4. The method for low-altitude intelligent network environment situational awareness based on machine learning according to claim 3, characterized in that, The specific implementation process of analyzing the environmental impact relationships between different low-altitude sensing areas, combining the spatial distance influence weight with the state transition law to calculate the degree of propagation of environmental changes between areas, and forming a low-altitude situation evolution trajectory covering multiple areas includes: Regional correlation weights are established based on the spatial distance between different regions. The spatial correlation weights are combined with the state transition probability to calculate the degree of influence of the propagation of situational changes between regions, so as to determine the possible degree of impact of environmental changes in a certain region on the surrounding regions. When environmental changes occur in a certain area, the surrounding affected areas are determined based on spatial correlation weights, thereby increasing the attention paid to future changes in the state of the affected areas and forming a regional-level situation evolution set.

5. The low-altitude intelligent network environment situational awareness method based on machine learning according to claim 3, characterized in that, The reliability of the generated situation evolution trajectory is evaluated by comprehensively considering the matching degree of state transition, spatial propagation reliability, and prediction error. When the trajectory reliability is lower than a set threshold, the specific implementation process of dynamically correcting the evolution trajectory by adjusting model parameters, updating historical samples, and enhancing the influence of real-time data includes: The reliability of the trajectory is calculated based on the degree of state transition matching, spatial propagation reliability, and prediction error. The reliability of the trajectory is determined by a weighted comprehensive analysis of historical state transition matching factors, spatial propagation factors, and prediction error factors. The weights of historical state transition matching factors, spatial propagation factors, and prediction error factors are determined based on historical operating data. A threshold for judging trajectory credibility is set. When the trajectory credibility meets the reliability requirements, the current situation evolution trajectory is directly used for subsequent situation analysis. When the trajectory credibility does not meet the reliability requirements, it is determined that the current trajectory has uncertainty and is dynamically adjusted. The dynamic adjustment methods include increasing the proportion of influence from real-time environmental status data, reducing the influence of abnormal historical samples, re-matching similar historical evolution trajectories, and recalculating the probability of future state migration. After adjustment, the final low-altitude situation evolution trajectory is formed.

6. The low-altitude intelligent network environment situational awareness method based on machine learning according to claim 5, characterized in that, Step S4 specifically includes: Based on the regional situation evolution trajectory, information on environmental state changes within multiple future time windows is extracted, the amount of change between consecutive states is calculated, and regional active perception demand indicators are constructed by combining state transition probability and trajectory reliability. By setting a preset demand threshold, it is determined whether different regions need to enter an active perception state, and the future environmental change trend is converted into a basis for adjusting perception resources. Based on the regional active perception demand indicators, the collection cycle of perception nodes and edge computing tasks in the low-altitude intelligent network are dynamically adjusted. By normalizing the perception indicator requirements of different regions, the proportion of perception resource allocation is determined, and the data update frequency and computing task priority of the corresponding regions are adjusted so that computing resources can serve the regions with future state change needs. The predicted environmental state is corrected in real time by using dynamically adjusted sensing data. The error between the predicted state and the actual collected state is calculated, and the correction method of the prediction result is determined according to the error threshold. The predicted data and real-time data are then integrated to form the corrected environmental state. The environmental state after feedback correction, the active perception demand indicators, and the trajectory credibility information are formed into closed-loop update data and fed back to the machine learning environmental state evolution model for continuous optimization. By evaluating the prediction accuracy, resource utilization, and model operation stability, the subsequent perception strategy, collection cycle, and computing task allocation method are updated.

7. The method for low-altitude intelligent network environment situational awareness based on machine learning according to claim 6, characterized in that, The specific steps for dynamically adjusting the data collection cycle of sensing nodes and edge computing tasks in the low-altitude intelligent network based on regional active sensing demand indicators include: The future environmental state changes of each region are analyzed by using the situation evolution trajectory, the amount of environmental state change between consecutive prediction time windows is calculated, and the future change intensity index is constructed by combining the state changes of multiple prediction time windows. The time weight of the recent prediction time window is higher than that of the long-term prediction time window. By combining state transition probability and trajectory reliability, a regional future perception demand index is constructed, and a preset perception demand threshold is set. When the regional future perception demand index reaches the active perception condition, the region is designated as a key data update region. The future sensing demand indicators of all low-altitude sensing areas are normalized to convert the sensing demand indicators of different areas into resource allocation criteria under a unified scale. The data collection cycle of the corresponding area is adjusted according to the sensing resource allocation ratio. When the future sensing demand indicator of an area is higher than the threshold, the data collection cycle of that area is shortened. When the future sensing demand indicator of an area is not higher than the threshold, the basic data collection cycle is maintained. The computational task weights for the corresponding regions are determined based on the regional perception requirement indicators. The tasks are then sorted for execution on the edge computing nodes according to the computational task weights. For regions where the computational task weights reach the preset task scheduling conditions, environmental state update tasks, state transition analysis tasks, and situation prediction tasks are executed.

8. A machine learning-based low-altitude intelligent network environment situational awareness system, used to execute the machine learning-based low-altitude intelligent network environment situational awareness method according to any one of claims 1 to 7, characterized in that, include: The multi-source heterogeneous environment data acquisition and state primitive construction module is used to collect, synchronize, and fuse multiple heterogeneous data on flight activities, communication status, meteorological disturbances, and space occupancy in the low-altitude intelligent network, and construct a unified expression of environmental state primitives. The Environmental State Evolution Learning and Situation Prediction Module is used to build machine learning analysis models based on environmental state primitives, explore the changing patterns of low-altitude environmental states, analyze the correlation between different environmental parameters, and generate the future state evolution results of the region. The situation evolution trajectory generation and state transition analysis module is used to establish regional situation evolution trajectories based on historical state information and prediction results, calculate environmental state transition relationships and trajectory reliability, and form a description of low-altitude environmental situation changes for future time windows. The active perception and control module is used to calculate the regional perception requirements based on the situational evolution results, dynamically adjust the data acquisition strategy and edge computing task allocation, and use real-time feedback data to correct the prediction results, thereby realizing the continuous optimization of the machine learning environment state evolution model and perception strategy.

Images